IJACSA Volume 15 Issue 11 - thesai.org

IJACSA Volume 15 Issue 11

Copyright Statement: This is an open access publication licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, even commercially as long as the original work is properly cited.

View Full Issue

Paper 1: Predicting Cervical Cancer Based on Behavioral Risk Factors

Abstract: Machine learning (ML) based predictive models are increasingly used in various fields due to their ability to find patterns and interpret complex relationships between variables in an extensive dataset. However, getting a comprehensive dataset is challenging in the field of medicine for rare or emerging infections. Therefore, developing a robust methodology and selecting ML classifiers that can still make compelling predictions even with smaller and imbalanced datasets is essential to defend against emerging threat or infections. This paper uses behavioral risk factors to predict cervical cancer risk. To create a robust technique, we intentionally selected a smaller imbalanced dataset and applied Adaptive Synthetic (ADASYN) sampling and hyper-parameter tuning to enhance the predictive performance. In this work, hyperparameter tuning, evaluated through 3-fold cross-validation, is employed to optimize the performance of the Random Forest, XGBoost, and Voting Classifier models. The results demonstrated high classification performance, with all models achieving an accuracy of 97.12%. Confusion matrix analysis further revealed the models’ robustness in identifying cervical cancer cases with minimal misclassification. A comparison with previous work confirmed the superiority of our approach, showcasing improved accuracy and precision. This study demonstrates the potential of ML models for early screening and risk assessment, even when working with limited datasets.

Author 1: Rakeshkumar Mahto

Author 2: Kanika Sood

Keywords: Cervical cancer; random forest; voting classifier; Adaptive Synthetic Sampling (ADASYN); predictive model

Paper 2: Comparative Analysis of Machine Learning Models for Forecasting Infectious Disease Spread

Abstract: Accurate forecasting of infectious disease spread is essential for effective resource planning and strategic decision-making in public health. This study provides a comprehensive evaluation of various machine learning models, from traditional statistical approaches to advanced deep learning techniques, for forecasting disease outbreak dynamics. Focusing on daily positive cases and daily deaths—key indicators despite potential reporting inconsistencies—our analysis aims to identify the most effective models across different algorithm families. By adapting non-time series methods with temporal factors and enriching time series models with exogenous variables, we enhance model suitability for the data’s time-dependent nature. Using India as a case study due to its significant early pandemic spread, we evaluate models through metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Median Squared Error (MEME), and Mean Squared Log Error (MSLE). The models tested include Linear Regression, Elastic Net, Random Forest, XGBoost, and Simple Exponential Smoothing, among others. Results indicate that the Random Forest Regressor outperforms other methods in terms of prediction accuracy across most metrics. Notably, findings suggest that simpler models can sometimes match or even exceed the reliability of more complex approaches. However, limitations include model sensitivity to data quality and the lack of real-time adaptability, which may affect performance in rapidly evolving outbreak situations. These insights have critical implications for public health policy and resource allocation in managing infectious disease outbreaks.

Author 1: Praveen Damacharla

Author 2: Venkata Akhil Kumar Gummadi

Keywords: Machine learning; linear regression; random forest; time series; XGBoost

Paper 3: Augmented Reality in Education: Revolutionizing Teaching and Learning Practices – State-of-the-Art

Abstract: The evolution and contemporary applications of instructional technology, particularly the transformative impact of Augmented Reality (AR) in education, are comprehensively explored in this study. Tracing the journey from early visual aids to sophisticated AR, the aim is to highlight continuous efforts to enhance educational experiences. The effectiveness of AR in increasing student engagement, comprehension, and personalized learning across various disciplines is critically assessed, revealing its potential to transform abstract concepts into tangible experiences. Additionally, challenges in AR adoption, such as technological constraints, the necessity for comprehensive educator training, and strategic curriculum integration, are discussed. The objective here is to identify research gaps, emphasizing the need for standardized evaluation methods, larger sample sizes, and long-term impact studies to fully understand AR’s potential. This exploration aims to provide a comprehensive understanding of AR’s capability to revolutionize education and to identify pathways for future research and development in this dynamic field.

Author 1: Samer Alhebaishi

Author 2: Richard Stone

Keywords: Augmented reality; education; instructional technology; technology integration; student engagement; teacher training

Paper 4: The Future of Mainframe IDMS: Leveraging Artificial Intelligence for Modernization and Efficiency

Abstract: IDMS (Integrated Database Management System) has long been a backbone for mission-critical systems in finance, healthcare, and government sectors. However, the rigid architecture of legacy systems poses challenges in scalability, flexibility, and integration with modern technologies. This paper explores IDMS modernization using Artificial Intelligence (AI), with a focus on predictive maintenance, query optimization, and cloud integration. Through real-world implementations, the integration of AI-driven solutions has shown transformative potential: query response times were reduced by 25%, unscheduled downtime decreased by 30%, and system scalability improved by accommodating a 40% increase in traffic without degradation. By leveraging AI-powered automation and modern cloud infrastructures, IDMS can achieve database optimization and real-time operational efficiency. This work highlights how AI ensures the relevance and competitiveness of IDMS, enabling it to meet the demands of modern legacy systems and ensuring its sustained role in critical business operations.

Author 1: Vasanthi Govindaraj

Keywords: IDMS modernization; artificial intelligence in legacy systems; mainframe database optimization; predictive maintenance; cloud integration; AI-driven query optimization

Paper 5: The Future of IoT Security in Saudi Arabian Start-Ups: A Position Paper

Abstract: This research explores the intricacies of implementing and securing Internet of Things (IoT) technology in Saudi Arabian startups. In the middle of Saudi Arabia's ambitious pursuit of social and economic progress via IoT breakthroughs, entrepreneurs have emerged as critical participants grappling with serious computing security challenges. This study conducts a thorough examination of the cybersecurity risks associated with start-up in Saudi Arabia’s IoT applications, technologies and innovations by reviewing a wide range of publications. The key objectives are to identify the main cybersecurity risks, analyze the impact of IoT device networking on privacy and security, and propose strategies to mitigate these threats. Furthermore, the study stresses the importance of funding up-to-date security technologies, cooperation with the cyber experts, and shifting towards the cloud-based security. Also, the study identifies the importance of cybersecurity education and training to enhance the defensive mechanisms of the startups against cyber threats. This study provides novel insights by identifying the distinct cybersecurity obstacles encountered by IoT-enabled businesses in Saudi Arabia and proposing a complete framework to enhance their security architecture. Robust cybersecurity policies are vital for both unleashing the transformative potential of IoT for startups and guiding Saudi Arabia towards its objective of being a worldwide leader in IoT. This paper advocates for a cooperative strategy that involves policymakers, industry stakeholders, and entrepreneurs to prioritize and allocate resources towards a safe and robust IoT ecosystem. This would help promote economic development and innovation in the country.

Author 1: Safar Albaqami

Author 2: Maziar Nekovee

Author 3: Imran Khan

Keywords: IoT; Saudi Arabia; start-ups; computing security challenges; technology; innovation; cyber threats

Paper 6: A Low-Cost IoT Sensor for Indoor Monitoring with Prediction-Based Data Collection

Abstract: The proliferation of Internet of Things technologies has revolutionized the landscape of indoor environmental monitoring, offering opportunities to enhance comfort, health, and energy efficiency. This paper presents the development and implementation of a low-cost IoT sensor system designed for indoor monitoring with a Machine Learning-driven prediction-based data collection approach. Leveraging deep learning algorithms, the IoT device predicts significant environmental changes and dynamically adjusts the data collection frequency to optimize energy consumption and data transmission. Experimental results demonstrate the system’s ability to accurately predict environ-mental variations, resulting in a reduction in data transmission and power usage up to 96% without compromising the monitoring quality. The findings highlight the potential of prediction-based data collection as a viable solution for sustainable and effective indoor environment monitoring on low-cost IoT devices.

Author 1: Paolo Capellacci

Author 2: Lorenzo Calisti

Author 3: Emanuele Lattanzi

Keywords: IoT; indoor monitoring; prediction-based data collection; deep-learning

Paper 7: Reliable Logistic Regression for Credit Card Fraud Detection

Abstract: Credit card fraud poses a significant threat to financial institutions and consumers worldwide, necessitating robust and reliable detection methods. Traditional classification models often struggle with the challenges of imbalanced datasets, noise, and outliers inherent in transaction data. This paper introduces a novel fraud detection approach based on a discrete non-additive integral with respect to a non-monotonic set function. This method not only enhances classification performance but also provides an interval-valued output that serves as an index of reliability for each prediction. The width of this interval correlates with the prediction error, offering valuable insights into the confidence of the classification results. Such an index is crucial in high-stakes scenarios where misclassifications can have severe consequences. The model is validated through extensive experiments on credit card transaction datasets, demonstrating its effectiveness in handling imbalanced data and its superiority over traditional models in terms of accuracy and reliability assessment. However, potential challenges such as increased computational complexity and the need for careful parameter tuning may affect scalability and real-time implementation. Addressing these challenges could further enhance the practical applicability of the proposed method in fraud detection systems.

Author 1: Yassine Hmidy

Author 2: Mouna Ben Mabrouk

Keywords: Credit card fraud; fraud detection; computational complexity

Paper 8: AI Ethical Framework: A Government-Centric Tool Using Generative AI

Abstract: Artificial Intelligence (AI) is transforming industries and societies globally. To fully harness this advancement, it is crucial for countries to integrate AI across different domains. Moral relativism in AI ethics suggests that as ethical norms vary significantly across societies, frameworks guiding AI development should be context-specific, reflecting the values, norms, and beliefs of the cultures where these technologies are deployed. To address this challenge, we introduce an intuitive, generative AI based solution that could help governments establish local ethical principles for AI software and ensure adherence to these standards. We propose two web applications: one for government use and another for software developers. The government-centric application dynamically calibrates ethical weights across domains such as the economy, education, and healthcare according to sociocultural context. By using LLMs, this application enables the creation of a tailored ethical blueprint for each domain or context, helping each country or region better define its core values. For developers, we propose a diagnostic application that actively checks software, assessing its alignment with the ethical principles established by the government. This feedback allows developers to recalibrate their AI applications, ensuring they are both efficient and ethically suitable for the intended area of use. In summary, this paper presents a tool utilizing LLMs to adapt software development to the ethical and cultural principles of a specific society.

Author 1: Lalla Aicha Kone

Author 2: Anna Ouskova Leonteva

Author 3: Mamadou Tourad Diallo

Author 4: Ahmedou Haouba

Author 5: Pierre COLLET

Keywords: AI ethics; Gen AI; LLMs; moral relativism; ethical norms; adaptive ethical framework

Paper 9: Simulation-Based Analysis of Evacuation Information Sharing Systems Using Geographical Data

Abstract: In this study, we developed an agent-based model (ABM) to simulate and improve evacuation rates during flood disasters. Utilizing the “Evacuate Now Button”, a previously proposed system for sharing real-time evacuation rates among residents, our experimental findings demonstrate a significant enhancement in evacuation behavior through this system. Simulations were conducted using geographical data from Nobeoka City, Miyazaki Prefecture, and Toyohashi City, Aichi Prefecture. Results showed that the “Evacuate Now Button” increased evacuation rates from a few percent to approximately 78% in Nobeoka City and 90% in Toyohashi City. We also investigated the effect of varying the range for calculating evacuation rates and the accuracy of the evacuation information shared with residents. It was found that larger calculation ranges led to higher final evacuation rates, while smaller ranges resulted in a quicker initial increase in evacuation behavior. These findings provide valuable insights for enhancing evacuation strategies and disaster preparedness in regions prone to floods.

Author 1: Tatsuki Fukuda

Keywords: Evacuation; flood disaster; evacuation rate; agent-based model; evacuate now button

Paper 10: Application of Unbalanced Optimal Transport in Healthcare

Abstract: Optimal Transport (OT) is a powerful tool widely used in healthcare applications, but its high computational cost and sensitivity to data changes make it less practical for resource-constrained settings. These limitations also contribute to increased environmental impact due to higher CO2 emissions from computing. To address these challenges, we explore Unbalanced Optimal Transport (UOT), a variation of OT that is both computationally efficient and more robust to data variability. We apply UOT to two healthcare scenarios: independence testing on breast cancer data and modeling heart rate variability (HRV). Our experiments show that UOT not only reduces computational costs but also delivers reliable results, making it a practical alternative to OT for socially impactful applications.

Author 1: Qui Phu Pham

Author 2: Nghia Thu Truong

Author 3: Hoang-Hiep Nguyen-Mau

Author 4: Cuong Nguyen

Author 5: Mai Ngoc Tran

Author 6: Dung Luong

Keywords: Optimal transport; unbalanced optimal transport; healthcare

Paper 11: Fuzzy Logic-Driven Machine Learning Algorithms for Improved Early Disease Diagnosis

Abstract: Early disease diagnosis is critical in improving patient outcomes, reducing healthcare costs, and preferably timely intervention. Unfortunately, the algorithms used in conventional diagnostic technology have difficulties dealing with uncertain and imprecise medical data, which may result in either delay or misdiagnosis. This paper describes the combined framework of fuzzy logic and machine learning algorithms to improve the accuracy and reliability of early disease diagnosis. Fuzzy logic addresses imprecision in patient symptoms and variability in clinical data, while machine learning algorithms provide data analytical and predictive capabilities. The proposed system enhances the abilities and complements rule-based reasoning with a predictive model to handle imprecise inputs and deliver accurate disease risk estimation. An experimental analysis of the medical datasets of heart disease, diabetes, and cancer reveals that the proposed method enhances the accuracy, precision, and ultimately robustness of a conventional diagnostic system.

Author 1: Leena Arya

Author 2: Narasimha Swamy Lavudiya

Author 3: G Sateesh

Author 4: Harish Padmanaban

Author 5: B. V. Srinivasulu

Author 6: Ravi Rastogi

Keywords: Decision trees; Fuzzy Inference System (FIS); heart disease diagnosis; neural networks; Support Vector Machine (SVM)

Paper 12: Automatic Detection of Lumbar Spine Disc Herniation

Abstract: Advanced deep-learning approaches have set new standards for computer vision and pattern recognition. However, the complexity of medical images frequently impedes the creation of high-quality ground truth data. In this article, we offer a method for autonomously generating ground truth data from MRI images using instance segmentation, with a novel confidence and consistency metric to assess data quality. We employ an artificial intelligence-based system to annotate regions of interest in MRI images, leveraging Mask R-CNN—a deep neural network architecture with a mean average precision of 98% for localising and identifying discs. Subsequently, the region of interest is classified with an accuracy of 70%. Our approach facilitates radiologists by automating the detection of regions of interest in MRI images, leading to more efficient and reliable diagnoses with assured quality data. This research made significant advances by developing an automated system for medical image segmentation and implementing cutting-edge neural network technologies.

Author 1: Mohammed Al Masarweh

Author 2: Olukola Oluseyi

Author 3: Ala Alkafri

Author 4: Hiba Alsmadi

Author 5: Tariq Alwadan

Keywords: Lumbar Disc Herniation; MASK-RCNN; computer vision; artificial intelligence; MR Images

Paper 13: AI-Powered AOP: Enhancing Runtime Monitoring with Large Language Models and Statistical Learning

Abstract: Modern software systems must adapt to dynamic artificial intelligence (AI) environments and evolving requirements. Aspect-oriented programming (AOP) effectively isolates crosscutting concerns (CCs) into single modules called aspects, enhancing quality metrics, and simplifying testing. However, AOP implementation can lead to unexpected program outputs and behavior changes. This paper proposes an AI-enhanced, adaptive monitoring framework for validating program behaviors during aspect weaving that integrates AOP interfaces (AOPIs) with large language models (LLMs), i.e. GPT-Codex AI, to dynamically generate and optimize monitoring aspects and statistical models in realtime. This enables intelligent run-time analysis, adaptive model checking, and natural language (NL) interaction. We tested the framework on ten diverse Java classes from JHotdraw 7.6 by extracting context and numerical data and building a dataset for analysis. By dynamically refining aspects and models based on observed behavior, its results showed that the framework maintained the integrity of the Java OOP class while providing predictive insights into potential conflicts and optimizations. Results demonstrate the framework’s efficacy in detecting subtle behavioral changes induced by aspect weaving, with a 94% accuracy in identifying potential conflicts and a 37% reduction in false positives compared to traditional static analysis techniques. Furthermore, the integration of explainable AI provides developers with clear, actionable explanations for flagged behaviors through NL interfaces, enhancing interpretability and trust in the system.

Author 1: Anas AlSobeh

Author 2: Amani Shatnawi

Author 3: Bilal Al-Ahmad

Author 4: Alhan Aljmal

Author 5: Samer Khamaiseh

Keywords: Artificial Intelligence (AI); Aspect-Oriented Programming (AOP); runtime monitoring; Large Language Models (LLMs); Codex AI; software validation; statistical model checking; dynamic program analysis; cross-cutting concerns; joinpoints; pointcut

Paper 14: Optimizing Stroke Risk Prediction Using XGBoost and Deep Neural Networks

Abstract: Predicting brain strokes is inherently complex due to the multifaceted nature of brain health. Recent advancements in machine learning (ML) and deep learning (DL) algorithms have shown promise in forecasting stroke occurrences to a certain extent. This research paper explores the predictive potential of ML and DL models by utilizing a comprehensive dataset encom-passing diverse patient characteristics, including demographic factors, work culture, stress levels, lifestyle, and family history. Notably, this study incorporates 14 clinically significant attributes for prediction, surpassing the 10 attributes utilized by earlier researchers. To address existing limitations and enhance predictive accuracy, a novel ensemble model combining Deep Neural Networks (DNN) and Extreme Gradient Boosting (XGBoost) is proposed in this work. Also, a comparative analysis against individual DNN and XGBoost models, as well as Random Forest and Support Vector Machine (SVM) approaches are being done. The performance of the ensemble model is assessed using various metrics, including accuracy, precision, F1 score, and recall. The findings indicate that the DNN-XGBoost model exhibits superior predictive accuracy compared to standalone DNN and XGBoost models in identifying brain stroke occurrences.

Author 1: Renuka Agrawal

Author 2: Aaditya Ahire

Author 3: Dimple Mehta

Author 4: Preeti Hemnani

Author 5: Safa Hamdare

Keywords: DNN; XGBoost; stress level; stroke prediction

Paper 15: Financial Shifts, Ethical Dilemmas, and Investment Insights in Nursing Homes: A Pre- and Post-Pandemic Analysis

Abstract: The COVID-19 pandemic has significantly transformed the operational, financial, and ethical frameworks of nursing homes in the United States. This study offers a detailed analysis of the nursing home sector from 2015 to 2021, focusing on the financial viability and ethical standards before, during, and after the pandemic. The methodology employed a structured approach, including data federation, pre-processing, and trend analysis, using comprehensive datasets on Nursing Homes. The data was cleaned, standardized, and segmented into pre-pandemic (2015–2019), pandemic (2020), and post-pandemic (2021) periods to assess key trends and outcomes. The findings highlight how the pandemic exacerbated existing financial challenges, such as declining occupancy rates, increased operational costs, and reduced revenue streams, which led to closures and heightened investment activity in the sector. Government aid provided temporary stability, but long-term sustainability remains uncertain. Key factors affecting financial performance, including occupancy rates, net income, fines and penalties, and compliance with ethical standards such as vaccination rates and care quality, were analyzed. The study concludes that nursing home investments should be approached cautiously unless facilities meet specific financial and operational criteria, such as high occupancy rates, robust financial performance, low penalties, and strict adherence to ethical standards. Failure to meet these benchmarks may result in heightened financial and operational risks, making such facilities unsuitable for investment. This research offers a comprehensive framework for investors to evaluate nursing home opportunities in the post-pandemic landscape, providing insights into the intersection of financial performance, operational resilience, and ethical compliance.

Author 1: Amir El-Ghamry

Author 2: Ameera Ibrahim

Author 3: Noha Elfiky

Author 4: Safwat Hamad

Keywords: Component; COVID-19 impact; nursing home financial performance; post-pandemic investment; ethical standards in nursing homes

Paper 16: FusionSec-IoT: A Federated Learning-Based Intrusion Detection System for Enhancing Security in IoT Networks

Abstract: Internet of Things (IoT) has become one of the most significant technological advancements of the modern era, which has impacted multiple sectors in the way it provides communication between connected devices. However, this growth has led to security risks in the IoT devices especially when constructing resource-limited IoT networks that are easily attacked by hackers through methods like DDoS and data theft. Traditional IDS such as centralized IDS and single-view machine learning-based IDS are incapable of providing efficient solutions to these issues due to high communication cost, latency, and low attack detection rate for IDS. To address these challenges, this paper presents FusionSec-IoT, a decentralized IDS based on multi-view learning and federated learning for better detection performance and scalability in the IoT context. The results sows that the proposed technique performs better than the existing IDS methods with 08.3% accuracy as compared to classic IDS techniques centralized IDS (91.5%) and single-view federated learning (92.7%). The other performance metrics like shows a better performance as compared to traditional IDS methods. Thus, FusionSec-IoT detects a broad range of cyberattacks with the help of the employed complex machine learning algorithms such as Reinforcement Learning, Meta-Learning, and Hybrid Feature Selection using Particle Swarm Optimisation (PSO) and Genetic Algorithm (GA), and ensures data privacy is maintained. Moreover, Edge Computing and Graph Neural Networks (GNNs) guarantee fast identification of multi-device coordinated attacks, for instance, botnets. The above-discussed proposed system enhances the traditional IDS approaches in terms of high detection accuracy, better privacy, and scalability, making the proposed system a reliable solution to secure the complex and large-scale IoT networks.

Author 1: Jatinder Pal Singh

Author 2: Rafaqat Kazmi

Keywords: IoT security; Intrusion Detection System (IDS); federated learning; multi-view learning; cyberattack detection

Paper 17: Incorporating Local Texture Adversarial Branch and Hybrid Attention for Image Super-Resolution

Abstract: In the field of image Super-Resolution reconstruction (SR), traditional SR techniques such as regression-based methods and CNN-based models fail to retain texture details in the reconstructed images. Conversely, Generative Adversarial Networks (GANs) have significantly enhanced the visual quality of image reconstruction through their adversarial training architecture. However, existing GANs still exhibit limitations in capturing local details and efficiently utilizing features. To address these challenges, we have proposed a super-resolution reconstruction method leveraging local texture adversarial and hybrid attention mechanisms. Firstly, a Local Texture Sampling Module (LTSM) is designed to precisely locate small regions with strong texture features within an image, and a local discriminator then performs pixel-by-pixel evaluation on these regions to enhance local texture details. Secondly, a hybrid attention module is integrated into the generator’s residual module to improve feature utilization and representativeness. Finally, we conducted extensive experiments to validate the effectiveness of our method. The results demonstrate that our method surpasses other super-resolution reconstruction methods in terms of PSNR and SSIM on four benchmark datasets. Furthermore, our method visually generates high-resolution images with richer details and more realistic textures.

Author 1: Na Zhang

Author 2: Hanhao Yao

Author 3: Qingqi Zhang

Author 4: Xiaoan Bao

Author 5: Biao Wu

Author 6: Xiaomei Tu

Keywords: Super-resolution reconstruction; generative adversarial network; hybrid attention; local texture sampling

Paper 18: Image Restoration of Landscape Design Based on DCGAN Optimization Algorithm

Abstract: To enhance the quality and effectiveness of image restoration in landscape design, this study optimizes the existing methods for low efficiency and incomplete feature extraction in processing high-resolution and detail rich landscape design images. Firstly, based on the traditional generative adversarial network (GAN), a novel deep convolutional generative adversarial network (DCGAN) model is proposed. Subsequently, the model's ability to extract detailed features was enhanced by integrating dense connected networks (DenseNet) and compressed excitation networks (SENet) into the network architecture. An improved DCGAN is designed for the restoration of landscape design images. According to the results, the optimized model had a restoration precision and repair recall rate of 0.97 in benchmark performance testing, which was significantly better than traditional deep convolutional generative adversarial network models. In practical applications, the model had an average accuracy of over 97% in repairing four different styles of landscape images, with an average repair time as low as 0.06s. From this, it can be seen that the designed model can provide a more efficient technical means for the restoration and digital preservation of landscape design images.

Author 1: Wenjun Zhang

Keywords: Deep convolutional generative adversarial network; image; restoration; landscape architecture; squeeze-and-excitation network; dense convolutional network

Paper 19: Tennis Action Evaluation Model Based on Weighted Counter Clockwise Rotation Angle Similarity Measurement Method

Abstract: In order to intelligently analyze tennis movements and improve evaluation efficiency, a counter clockwise rotation angle of limbs is proposed to solve the direction problem of tennis movements. A dynamic time regularization algorithm is optimized by combining global time weighting and adjacent frame weighting. The results indicated that the proposed counter clockwise rotation angle feature of limbs could effectively represent changes in limb direction and clearly distinguish action postures. The average accuracy of this method in action classification on the Tennis Stroke Dataset was 97.60%. In the action evaluation mode, the average frame rate of the client was between 17.35FPS and 17.49FPS, and the overall average frame rate was about 17.40FPS. The server exhibits higher efficiency in action processing and evaluation, which can process video frames faster. It is more efficient in processing data capabilities and utilizing data resources. This indicates that the performance of the system is relatively consistent in different modes and has stability. The optimized method has a higher generalization ability in recognizing non-tennis movements on different datasets. When dealing with fine movements, the optimized method performs excellently and can better capture subtle differences in the movements. Meanwhile, this enhances the real-time performance of the system, making it suitable for evaluating tennis movements in practical application scenarios. This provides a new technical path for analyzing tennis movements and also serves as a reference for evaluating movements in other sports.

Author 1: Danni Jiang

Author 2: Ge Liu

Keywords: Action evaluation; counter clockwise rotation angle; weighting; dynamic time warping; tennis

Paper 20: Internet of Things User Behavior Analysis Model Based on Improved RNN

Abstract: Currently, there are issues with low efficiency and outdated Internet of Things resource allocation. To study real Internet of Things user behavior data, a Bayesian optimization algorithm is used to automatically select hyperparameter combinations and construct an Internet of Things user behavior analysis model based on long short-term memory. The results showed that the prediction accuracy of the model reached 96.8% and 97.53% on the training and validation sets, while in the set 50 maximum iterations, the model achieved 80.78% on the test set. In comparing the performance between the research model and the traditional recurrent network model, it was found that the optimal prediction accuracy of the research model was 80.78%, which was better than the comparison model. The application results of the research model in short-term power load forecasting also indicated that the prediction accuracy of the Internet of Things user behavior analysis model based on the improved recurrent network has reached a good level, far superior to the comparative model. The results have important application value for allocating energy and resources in Internet of Things systems.

Author 1: Keling Bi

Keywords: Internet of Things; user behaviors; recurrent neural network; Bayesian optimization; long short-term memory; hyperparameter

Paper 21: Network Security Based on Improved Genetic Algorithm and Weighted Error Back-Propagation Algorithm

Abstract: In order to solve the problem of feature selection and local optimal solution in the field of network security, a network security protection model based on improved genetic algorithm and weighted error back-propagation algorithm is proposed. The model combines the dynamic error weight and adaptive learning rate of the weighted error back-propagation algorithm to improve the learning ability of the model in dealing with classification imbalance and dynamic attack mode. In addition, the global search capability of genetic algorithm is utilized to optimize the feature selection process and automatically adjust the hyperparameter settings. The experimental results show that the proposed model has an average accuracy of 96.7%, a recall rate of 93.3% and an F1 value of 0.91 on the CIC-IDS-2017 dataset, which has significant advantages over traditional detection methods. In many experiments, the accuracy of normal data is up to 99.97%, the accuracy of known abnormal behavior data is 99.31%, and the accuracy of unknown abnormal behavior data is 98.13%. These results show that this method has high efficiency and reliability when dealing with complex network traffic, and provides a new idea and method for network security protection research.

Author 1: Junjuan Liang

Keywords: Genetic algorithm; weighted error back-propagation; multiple strategies; network security

Paper 22: Q-FuzzyNet: A Quantum Inspired QIF-RNN and SA-FPA Optimizer for Intrusion Detection and Mitigation in Intelligent Connected Vehicles

Abstract: In the evolving landscape of Intelligent Connected Vehicles (ICVs), ensuring cybersecurity is crucial due to the increasing number of cyber threats. Besides, challenges like data breaches, unauthorized access, and hacking attempts are prevalent due to the interconnected nature of ICVs. Several methods have been proposed to secure ICVs; however, accurate intrusion detection remains a challenging task yet to be fully achieved. For this reason, this paper proposes a comprehensive intrusion detection scheme denoted a Q-FuzzyNet, which is specifically tailored to safeguard ICV networks using Deep Learning (DL) approaches. This Q-FuzzyNet approach consists of five phases: (i) Data Collection (ii) Data Pre-processing (iii) Feature extraction (iv) Dimensionality Reduction and (v) Intrusion Detection and Mitigation. Initially, the raw data are gathered from the CICIoV2024 dataset. The collected data are pre-processed via Mean Imputation (MI) for data cleaning. Then, significant features are extracted through higher-order statistical features, Proposed Improved Mutual Information (IMI), Correlation, and Entropy approaches. Subsequently, the dimensionality is reduced via new Improved Linear Discriminant Analysis (ILDA). Ultimately, the data are classified (attacker/Normal) via the Meta-Heuristic Quantum-Inspired Fuzzy-Recursive Neural Network (QIF-RNN) model by combining the Quantum Neural Network (QNN), Recurrent Neural Network (RNN), and Fuzzy logic. The membership function of fuzzy logic is optimized via the new Self Adaptive-Flower Pollination Algorithm (SA-FPA). The identified attackers are mitigated from the network using the Policy Gradient Method. The acquired outcomes from Q-FuzzyNet are validated in terms of Accuracy, Precision, Sensitivity, and F1-score, as well. The highest accuracy of 98.6% has been recorded by the proposed model.

Author 1: Abdullah Alenizi

Keywords: Cybersecurity; intelligent connected vehicles; artificial intelligence; quantum neural network; recurrent neural network; flower pollination algorithm

Paper 23: Predicting Learners’ Academic Progression Using Subspace Clique Model in Multidimensional Data

Abstract: Subspace clustering examines the traditional clustering techniques that have previously been considered the best approaches to clustering data. This study uses a subspace clustering approach to predict learners' academic progress over time. Using the subspace clustering method, a model was developed that improves the classic Clique by optimizing clustering performance and addresses the clustering challenges posed by inaccuracies due to additional data size and increased dimensionality. The study used an experimental design that included data validation and training to predict students' academic progress. Clustering evaluation metrics including accuracy, precision, and recall measures were identified. The optimized model recorded a better performance index with 98.90% accuracy, 98.50% precision, and 98.50% recall which directly shows the efficiency of the optimized model in predicting learning academic progress through clustering. In this regard, conclusions are drawn for an alternative approach to predictive modeling through cluster analysis, so that educational institutions have a better opportunity to manage learners by ensuring adequate preparation in terms of resources, policies and knowledge. It highlights career guidance for learners based on their academic progress. The result validates the suitability of the model for clustering multidimensional data.

Author 1: Oyugi Odhiambo James

Author 2: Waweru Mwangi

Author 3: Kennedy Ogada

Keywords: Subspace clustering; clique model; academic progression; multidimensional data; feature engineering; cross validation and principal component analysis

Paper 24: Enhanced TODIM-TOPSIS Framework for Interior Design Quality Evaluation in Public Spaces Under Hesitant Fuzzy Sets

Abstract: The evaluation of interior landscape design in public spaces involves several aspects, including aesthetics, functionality, sustainability, and user experience. Aesthetic evaluation focuses on the visual appeal and stylistic consistency of the design. Functionality considers the practicality and convenience of the space layout. Sustainability evaluates the environmental friendliness of materials and energy efficiency of the design. Additionally, user experience assessment gathers feedback to gauge comfort and satisfaction. These evaluation criteria help designers optimize spaces to be both attractive and practical while meeting user needs. The interior design quality evaluation in public spaces is multiple-attribute decision-making (MADM) problem. Recently, the TODIM and TOPSIS methods have been applied to address MADM challenges. Hesitant fuzzy sets (HFSs) are used to represent uncertain information in the evaluation of interior landscape design in public spaces. In this study, we developed a hesitant fuzzy TODIM-TOPSIS (HF-TODIM-TOPSIS) approach to tackle Multiple Attribute Decision Making (MADM) issues within the context of HFSs. A numerical case study focused on the interior design quality evaluation in public spaces demonstrates the validity of this approach. The primary contributions of this paper include: (1) Extending the TODIM and TOPSIS approaches to incorporate HFSs; (2) Utilizing information entropy to determine weight values under HFSs; (3) Establishing the HF-TODIM-TOPSIS method for managing MADM in the presence of HFSs; (4) Conducting algorithmic analysis and comparative studies based on a numerical example to assess the practicality and effectiveness of the HF-TODIM-TOPSIS approach.

Author 1: Lu Peng

Keywords: Multiple-attribute decision-making (MADM); hesitant fuzzy sets (HFSs); TODIM; TOPSIS; design quality evaluation

Paper 25: ARO-CapsNet: A Novel Method for Evaluating User Experience in Immersive VR Furniture Design

Abstract: Immersive virtual reality (VR) technology has become an essential tool in enhancing user experience across industries, particularly in furniture design. With the ability to provide realistic, interactive, and immersive environments, it significantly improves user engagement and decision-making in product design. However, existing analysis methods lack precision in evaluating user experience within VR environments. This study aims to develop a more accurate and efficient model for analyzing the application of immersive VR in future furniture design. By integrating the Artificial Rabbit Optimization (ARO) algorithm with Capsule Networks (CapsNet), this research enhances the evaluation of user experience in immersive VR environments. The proposed method uses the ARO algorithm to optimize the parameters of CapsNet, which maps the relationship between the analysis indicators of furniture design and user experience. This model is tested against traditional methods such as CNN and CapsNet alone. The analysis focuses on key factors such as visual elements, interaction, and system performance, with performance metrics like root mean square error (RMSE) and R² value used for evaluation. Experimental results show that the ARO-CapsNet model achieves a RMSE of 0.17 and an R² value of 0.988, outperforming both CNN and CapsNet in terms of accuracy and efficiency. Additionally, the proposed model improves the immersive VR system's ability to deliver accurate user experience evaluations, making it a superior method for analyzing future furniture design applications. The integration of the ARO algorithm with CapsNet significantly enhances the precision of immersive VR user experience evaluations in furniture design. The ARO-CapsNet model not only improves evaluation accuracy but also increases system efficiency, providing a robust framework for future applications of VR in product design.

Author 1: Yin Luo

Author 2: Jun Liu

Author 3: Li Zhang

Keywords: Immersive virtual reality; furniture design; application analysis; artificial rabbit optimisation algorithm

Paper 26: Application Pigeon Swarm Intelligent Optimisation BP Neural Network Algorithm in Railway Tunnel Construction

Abstract: Due to the uncertainty and complexity of the risk factors of the urban railway tunnel project to increase the difficulty of risk analysis, so that the traditional risk assessment methods can not accurately assess the construction risk of the urban railway tunnel project. Aiming at the problems of the existing risk assessment algorithms, the construction risk assessment method of an urban railway tunnel project based on intelligent optimisation algorithm and machine learning algorithm is proposed. Firstly, for the problem of construction risk identification and assessment of municipal railway tunnel project, a tunnel construction risk identification and assessment scheme using a combination of intelligent optimization algorithm and machine learning algorithm is designed, and the principles and functions of each module of the risk assessment system are introduced; then, for the problem of risk assessment construction, a risk assessment algorithm based on the swarm intelligent optimization algorithm to improve the BP neural network is proposed; secondly, relying on the Hangzhou Secondly, relying on Xinfeng Road underground passage close to cross the underground line 9 tunnel and the side through the Hanghai intercity tunnel project in Hangzhou, the effectiveness of the construction risk assessment algorithm is verified from monitoring data and numerical simulation, and the risk control scheme is proposed in turn. The experimental results show that the risk assessment algorithm proposed in this paper effectively solves the problem of construction risk assessment of the urban railway tunnel project, and improves the prediction performance of the risk assessment algorithm, and verifies that the risk control scheme meets the construction safety requirements.

Author 1: Feng Zhou

Author 2: Hong Ye

Author 3: Jie Song

Author 4: Hui Guo

Author 5: Peng Liu

Keywords: Municipal railway tunnel construction optimization; scenario risk assessment; machine learning; pigeon flock optimisation algorithm

Paper 27: A Data-Driven Deep Machine Learning Approach for Tunnel Deformation Risk Assessment

Abstract: The shallow overburden pipe jacking over operation tunnel construction project in chalk stratum has the risk of deformation of the soil layer and the existing tunnel, which increases the difficulty of pipe jacking over construction, and the risk assessment and control become the key technology for the safe and successful completion of the construction. Aiming at the problems of the current deformation risk assessment and control method, such as the assessment system is not comprehensive, systematic and objective enough, the prediction accuracy is not efficient enough, and there is a lack of quantitative analysis, etc., a deformation risk assessment and control method is proposed to combine the heuristic optimization algorithm of human behaviour and deep machine learning algorithm for pipe jacking up to and across operation tunnels on shallow overburden of chalky sand stratum. Firstly, by analyzing the construction process of pipe jacking tunnel, the deformation risk factors of the construction process and the deformation risk control scheme are given; then, a deformation risk assessment and control algorithm with improved deep limit learning machine is proposed by combining human heuristic optimization algorithm; finally, the proposed deformation assessment and control model is applied to the deformation risk assessment and control problem of pipe jacking over operation tunnel on shallow overburden of pulverised sand stratum, and a finite element computational model is used to construct the data. Finally, the proposed deformation assessment and control model is applied to the problem of deformation risk assessment and control in a tunnel with shallow overburden in chalky sand stratum by using finite element computational model to construct the data set, training the deformation risk assessment and control model, and using the monitoring data as the test set to validate the validity of the proposed model algorithm, and solving the problem of the poor prediction accuracy of the control algorithm for deformation risk assessment and control of a tunnel with shallow overburden in a tunnel with shallow overburden in chalky sand stratum.

Author 1: Fusheng Liu

Keywords: Pipe jacking up and over operational tunnel construction; tunnel deformation risk assessment; deep limit learning machine; hybrid leader optimisation algorithm; control strategy

Paper 28: Scalp Disorder Imaging: How Deep Learning and Explainable Artificial Intelligence are Revolutionizing Diagnosis and Treatment

Abstract: Scalp disorders, affecting millions worldwide, significantly impact both physical and mental health. Deep learning has emerged as a promising tool for automated diagnosis, but ensuring model transparency and reliability is crucial. This review explores the integration of explainable AI (XAI) techniques to enhance the interpretability of deep learning models in scalp disorder diagnosis. We analyzed recent studies employing deep learning models to classify scalp disorders from image data and used XAI methods to understand the models' decision-making processes and identify potential biases. While deep learning has shown promising results, challenges such as data quality and model interpretability persist. Future research should focus on expanding the capabilities of deep learning models for real-time detection and severity prediction, while addressing limitations in data diversity and ensuring the generalizability of models across different populations. The integration of XAI techniques is essential for fostering trust in AI-powered scalp disease diagnosis and facilitating their widespread adoption in clinical practice.

Author 1: Vinh Quang Tran

Author 2: Haewon Byeon

Keywords: Scalp disorders; artificial intelligence; explainable artificial intelligence; deep learning; interpretability

Paper 29: A Theoretical Framework of Extrinsic Feedback Evaluation in Football Training Based on Motion Templates Using Motion Capture

Abstract: Motion capture technology (MoCap) has emerged as a pivotal innovation, significantly impacting various sectors, including sports. In football training, MoCap is especially crucial for analyzing player movements with precision. Despite its potential, there remains a notable gap in the utilization of MoCap to create motion templates (MTs) that generate extrinsic feedback, particularly in football. This article proposes a comprehensive theoretical framework for evaluating extrinsic feedback in football training through MTs created using MoCap technology and Reverse-Gesture Description Language (R-GDL). The development of this framework involves several key steps: a literature review, acquaintance meetings, interviews, procedural approvals, experimentation, data conversion, MTs creation, and data evaluation. The framework integrates elements such as football players, MoCap systems, raw and processed data, MTs, evaluation processes, and extrinsic feedback models. The main purpose is to harness the full potential of MoCap technology, enhancing the evaluation and improvement of football training activities. By implementing this framework, we aim to revolutionize how football training is analyzed and optimized, providing coaches and players with invaluable insights and feedback.

Author 1: Amir Irfan Mazian

Author 2: Wan Rizhan

Author 3: Normala Rahim

Author 4: Muhammad D. Zakaria

Author 5: Mohd Sufian Mat Deris

Author 6: Fadzli Syed Abdullah

Author 7: Ahmad Rafi

Keywords: Motion capture; motion templates; football; extrinsic feedback; reverse-gesture description language

Paper 30: An Application of Graph Neural Network Model Design for Residential Building Layout Design

Abstract: In the current process of residential building layout design, there are problems such as low design efficiency, excessive manual intervention, and difficulty in meeting personalized needs. To address these issues, a residential building layout design method based on graph neural network model is proposed to improve the intelligence level of residential building layout design. Firstly, the residential building floor plan layout design data are transformed into graph data suitable for graph neural network model processing. Then, deep learning techniques are used to analyse and identify the spatial distribution characteristics of the main functional areas in the space. Finally, the trained graph neural network model is applied to the actual residential building floor plan layout design and compared with the traditional method. The experimental results show that compared with the traditional computer-aided design method, the residential building floor plan layout design and optimisation method improves the completeness of the design scheme by about 2.3%, the rationality by about 3.6%, the readability by about 1.9%, and the effectiveness by about 10.3%. The method improves the efficiency and accuracy of residential building floor plan layout design, helps to shorten the design cycle and reduce the design cost, and helps to promote technological progress and sustainable development in the field of architectural design.

Author 1: Shiyu Wang

Author 2: Ningbo Wang

Keywords: Residential building layout plan; deep learning; GNN model; space utilization rate; resident comfort level; quantum particle swarm algorithm; Node2vec algorithm

Paper 31: An Intelligent Transport System for Prediction of Urban Traffic Congestion Level

Abstract: Developing a resilient infrastructure is crucial for nation-building by supporting innovations and promoting sustainable growth. The Kingdom of Saudi Arabia is striving to achieve the Sustainable Development Goals (SDGs) set by the United Nations. Industry, Innovation, and Infrastructure (I3) are some of the strategic objectives of the Kingdom’s Vision 2030 par with the United Nations’ SDGs. The objective is focused to develop trade and transport networks for international, regional, and local connectivity with an investment of billions of dollars to establish a robust transport network and improve the existing one for enhancing road safety to reduce the costs of deaths and serious injuries. For this, a control center for automatic monitoring could be established for 24x7 monitoring of traffic violators; the key project has been named the National Center for Transportation Safety, apart from launching the “Rental Contracts” facility with the Naql portal. Moreover, the growing urban population is causing more vehicles on the roads leading to more traffic congestion which has become severe during peak hours in the major cities causing several other issues such as environmental pollution, high greenhouse gases (GHGs) including CO2 emissions, health risks to the citizen and residents, poor air quality, higher risks of road safety, more energy consumption, discomfort to the commuters, and wastage of time and other resources. Therefore, in this research, we propose an intelligent transport system (ITS) for predicting traffic congestion levels and assist commuters in taking alternative routes to avoid congestion. An intelligent model for predicting urban traffic congestion levels using XGBoost, Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) algorithms is developed. The comparative performance analysis of the techniques concerning the performance metrics: Mean Squared Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Error cost, Outlier sensitivity, and Model Complexity, demonstrate that the LSTM algorithm excels the other two algorithms.

Author 1: Mohammad Khalid Imam Rahmani

Author 2: Shahnawaz Khan

Author 3: Md Ezaz Ahmed

Author 4: Khurram Jawad

Keywords: Sustainable development goals; traffic congestion; traffic prediction; Gated Recurrent Unit; long short-term memory; intelligent transport system

Paper 32: SAM-PIE: SAM-Enabled Photovoltaic-Module Image Enhancement for Fault Inspection and Analysis Using ResNet50 and CNNs

Abstract: Different models have been developed for segmentation tasks, each with its uniqueness. Recently, the Segment Anything Model (SAM) was added to the pool of these models with expectations of addressing their weaknesses. SAM, although trained on a huge dataset for segmentation of anything, particularly images of natural source, produces suboptimal results when applied to segmentation of photovoltaic module image due to difference in semantic between photovoltaic module and natural images. In spite of the current suboptimal performance of SAM in segmentation of photovoltaic module images, it demonstrates detection and identification of thermal anomalies in photovoltaic module images that majorly contribute to power production loss. The implication of this is that, the task, the model, and the data corresponding to SAM are applicable to photovoltaic module image diagnosis. In this paper, we propose SAM-enabled photovoltaic-module image enhancement (SAM PIE) for fault inspection and analysis using ResNet50 and CNNs. SAM-PIE combines the strength of SAM for enhancement of the fault inspection and analysis procedure, for optimal performance of the proposed method. Experiments were performed on three thermal anomaly image datasets of photovoltaic modules to validate the performance of SAM-PIE for the classification tasks. The results obtained validates the potential capability of SAM-PIE to perform photovoltaic module image classification. The dataset is publicly and freely available for scientific community use at https://doi.org/10.17632/5ssmfpgrpc.1

Author 1: Rotimi-Williams Bello

Author 2: Pius A. Owolawi

Author 3: Etienne A. van Wyk

Author 4: Chunling Du

Keywords: Anomaly; convolution neural networks; crack; hotspot; photovoltaic; Residual Network-50; shading

Paper 33: Understanding Mental Health Content on Social Media and It’s Effect Towards Suicidal Ideation

Abstract: The study “Understanding Mental Health Content on Social Media and Its Effect Towards Suicidal Ideation” aims to detail the recognition of suicidal intent through social media, with a focus on the improvement and part of the machine learning (ML), deep learning (DL), and natural language processing (NLP). This review underscores the critical need for effective strategies to identify and support individuals with suicidal ideation, exploiting technological innovations in ML and DL to further suicide prevention efforts. The study details the application of these technologies in analyzing vast amounts of unstructured social media data to detect linguistic patterns, keywords, phrases, tones, and contextual cues associated with suicidal thoughts. It explores various ML and DL models like SVMs, CNNs, LSTM, neural networks, and their effectiveness in interpreting complex data patterns and emotional nuances within text data. The review discusses the potential of these technologies to serve as a life-saving tool by identifying at-risk individuals through their digital traces. Furthermore, it evaluates the real-world effectiveness, limitations, and ethical considerations of employing these technologies for suicide prevention, stressing the importance of responsible development and usage. The study aims to fill critical knowledge gaps by analyzing recent studies, methodologies, tools, and techniques in this field. It highlights the importance of synthesizing current literature to inform practical tools and suicide prevention efforts, guiding innovation in reliable, ethical systems for early intervention. This research synthesis evaluates the intersection of technology and mental health, advocating for the ethical and responsible application of ML, DL, and NLP to offer life-saving potential worldwide while addressing challenges like generalizability, biases, privacy, and the need for further research to ensure these technologies do not exacerbate existing inequities and harms.

Author 1: Mohaiminul Islam Bhuiyan

Author 2: Nur Shazwani Kamarudin

Author 3: Nur Hafieza Ismail

Keywords: Suicidal ideation detection; social media analysis; mental health; text analysis; machine learning

Paper 34: Classification of Liver Disease Using Conventional Tree-Based Machine Learning Approaches with Feature Prioritization Using a Heuristic Algorithm

Abstract: Liver disease ranks as one of the leading causes of mortality globally, often going undetected until advanced stages. This study aims to enhance early detection of liver disease by employing machine learning models that utilize key health indicators. Utilizing the Indian Liver Patient Dataset (ILPD) from the UCI repository, we developed a predictive model using the CatBoost algorithm, achieving an initial accuracy of 74%. To improve this, feature selection was performed using the Whale Optimization Algorithm (WOA) and Harris Hawk Optimization (HHO), which increased accuracy to 82% and 85% respectively. The methodology involved preprocessing to correct data imbalances and outlier removal through univariate and bivariate analyses. These optimizations highlight the critical features enhancing the model's predictive capability. The results indicate that integrating metaheuristic algorithms in feature selection significantly improves the accuracy of liver disease prediction models. Future research could explore the integration of additional datasets and machine learning models to further refine predictive capabilities and understand the underlying pathophysiology of liver diseases.

Author 1: Proloy Kumar Mondal

Author 2: Haewon Byeon

Keywords: Liver disease; classification; prediction; CatBoost algorithm; machine learning; optimization algorithm

Paper 35: Optimizing Deep Learning for Diabetic Retinopathy Diagnosis

Abstract: The detection of diabetic retinopathy traditionally requires the expertise of medical professionals, making manual detection both time- and labor-intensive. To address these challenges, numerous studies in recent years have proposed automatic detection methods for diabetic retinopathy. This research focuses on applying deep learning and image processing techniques to overcome the issue of performance degradation in classification models caused by imbalanced diabetic retinopathy datasets. It presents an efficient deep learning model aimed at assisting clinicians and medical teams in diagnosing diabetic retinopathy more effectively. In this study, image processing techniques, including image enhancement, brightness correction, and contrast adjustment, are employed as preprocessing steps for fundus images of diabetic retinopathy. A fusion technique combining color space conversion, contrast limited adaptive histogram equalization, multi-scale retinex with color restoration, and Gamma correction is applied to highlight retinal pathological features. Deep learning models such as ResNet50-V2, DenseNet121, Inception-V3, Xception, MobileNet-V2, and InceptionResNet-V2 were trained on the preprocessed datasets. For the APTOS-2019 dataset, DenseNet121 achieved the highest accuracy at 99% for detecting diabetic retinopathy. On the Messidor-2 dataset, InceptionResNet-V2 demonstrated the best performance, with an accuracy of 96%. The overall aim of this research is to develop a computer-aided diagnosis system for classifying diabetic retinopathy.

Author 1: Krit Sriporn

Author 2: Cheng-Fa Tsai

Author 3: Li-Jia Rong

Author 4: Paohsi Wang

Author 5: Tso-Yen Tsai

Author 6: Chih-Wen Chen

Keywords: Diabetic retinopathy; deep learning; image processing technologies; imbalanced image dataset; computer aided diagnosis

Paper 36: Assessing the Usability of M-Health Applications: A Comparison of Usability Testing, Heuristics Evaluation and Cognitive Walkthrough Methods

Abstract: Mobile health applications have increasingly become an important channel for providing services in the health sector. However, poor usability can be a major barrier for the rapid adoption of mobile services. The purpose of this study is to compare the relative performance of three usability evaluation methods, namely, usability testing, heuristics evaluation, and the cognitive walkthrough methods in determining the usability level of mobile health applications. The study also explores the relationship between the metrics of usability testing and the current level of mobile health applications in Saudi Arabia. An experimental approach has been used in this study, which gathered qualitative and quantitative data. The methods were used to assess two mobile health interfaces and were compared on the number, severity, and types of usability problems identified. Correlation tests were also carried out to examine areas of overlap between usability testing metrics. The heuristic evaluation found significantly greater numbers of usability problems than the other techniques. The usability testing method, however, detects problems of greater severity. There is also a significant correlation between the number of usability issues found and how long it takes to perform tasks in usability tests. Moreover, the level of usability of the Saudi applications tested is below expectation and in need of further improvement. Based on the study results, both usability testing and heuristic evaluation should be employed during the design process of mobile health applications for maximum effectiveness. Additionally, it is recommended that SUS questionnaires should not be the sole method of determining the usability level of mobile health applications.

Author 1: Obead Alhadreti

Keywords: Mobile health applications; usability; usability testing; heuristics evaluation; cognitive walkthrough

Paper 37: Automated Hydroponic Growth Simulation for Lettuce Using ARIMA and Prophet Models During Rainy Season in Indonesia

Abstract: Hydroponic farming particularly lettuce cultivation, is gaining popularity in Indonesia due to its economical use of water and space, as well as its short growing season. This study focuses on developing of an Automated Hydroponic Growth Simulation for Lettuce Using ARIMA and Prophet Models during the Rainy Season in Indonesia. We developed a simulation model for lettuce development in the Nutrient Film Technique (NFT) hydroponic system using data collected over four harvest periods during the rainy season in early 2024. Two machine learning models, ARIMA and Prophet, are tested to see which is more effective at forecasting lettuce growth. The Prophet model has the greatest results, with a Mean Absolute Error (MAE) of 1.475 and a Root Mean Square Error (RMSE) of 1.808. Based on this, the Prophet model is utilized to create a web application using Streamlit for real-time growth predictions. Future studies should include more data, particularly from the dry season, to increase model flexibility, as well as investigate the use of other crops and machine learning methods, including hybrid models, to improve forecasts.

Author 1: Lendy Rahmadi

Author 2: Hadiyanto

Author 3: Ridwan Sanjaya

Keywords: ARIMA; automated; growth; hydroponic; prophet; simulation

Paper 38: Improved Real-Time Smoke Detection Model Based on RT-DETR

Abstract: Fire remains a major threat to society and economic activities. Given the real-time demands of smoke detection, most research in deep learning has focused on Convolutional Neural Networks. The Real-Time Detection Transformer (RT-DETR) introduces a promising alternative for this task. This paper extends RT-DETR to address challenges such as morphological variations and interference in smoke detection by proposing the Realtime Smoke Detection Transformer (RS-DETR). RS-DETR uses smoke images with concentration data as input and employs a deformable attention module to manage morphological changes, enabling robust feature extraction. Additionally, a Cross-Scale Smoke Feature Fusion Module (CS-SFFM) is integrated to enhance detection accuracy for small and thin smoke targets through multi-scale feature resampling and fusion. To improve convergence speed and stability, Efficient Intersection over Union (EIoU) replaces Generalized Intersection over Union (GIoU) in feature scoring. The improved model achieves an average precision of 93.9% on a custom dataset, representing a 5.7% improvement over the original model, and demonstrates excellent performance across various detection scenarios.

Author 1: Yuanpan ZHENG

Author 2: Zeyuan HUANG

Author 3: Binbin CHEN

Author 4: Chao WANG

Author 5: Yu ZHANG

Keywords: RT-DETR; smoke detection; deformable convolution; multi-scale feature fusion; EIoU; image enhancement; dark channel

Paper 39: Predicting Stock Price Bubbles in China Using Machine Learning

Abstract: Financial bubbles have long been a focus of researchers, particularly due to the severe negative impacts following the bursting of financial bubbles. Therefore, the ability to effectively predict financial bubbles is of paramount importance. The aim of this study is to measure and predict the stock market price bubble in China from January 2015 to December 2023. To achieve this, we utilized the GSADF test, currently the most effective, to identify and measure the situation of the stock market price bubble in China. Subsequently, we selected inflation rate, consumer confidence index, stock yield, and price-earnings ratio as explanatory/predictive variables. Finally, four machine learning methods were employed to forecast the stock market price bubble in China. The results indicate that a price bubble occurred in the Chinese stock market during the first half of 2015, before the outbreak of the COVID-19 pandemic in China in January 2020. Furthermore, the comparison reveals that among the machine learning methods, logistic regression is the most suitable and effective for China, while other methods such as deep learning and decision trees also hold certain value.

Author 1: Yunxi Wang

Author 2: Tongjai Yampaka

Keywords: Stock price bubbles; machine learning; Chinese stock market

Paper 40: Multi-Factor Risk Assessment and Route Optimization for Safe Human Travel

Abstract: In the modern world, frequent travel has become a necessity, with vehicles being the primary mode of transportation. Ensuring human safety while traveling is paramount. To address this, it is essential to adopt a combination of numerous static and dynamic parameters to achieve optimal route design in today’s complex transportation systems. This study introduces a methodology titled 'Multi-Factor Risk Assessment and Route Optimization for Safe Human Travel', which consists of three stages: Route Optimization, Risk Factor Analysis, and Data Collection. To assess the safety of various routes, a combination of dynamic and static factors is considered. These include traffic, weather, and road conditions, as well as vehicle-related factors such as type, age, and the surrounding road environment. By analyzing simulated data, the technique identifies potential risks and optimizes travel paths accordingly. For segmented routes, risk factors are calculated using both static and dynamic parameters, ensuring a comprehensive safety assessment. Prioritizing user safety, the system dynamically adjusts routes to offer the most cost-effective and safest travel options. This study lays a robust foundation for intelligent transportation systems, aimed at ensuring safer travel for users across a range of scenarios.

Author 1: Thilagavathi T

Author 2: Subashini A

Keywords: Multi-factor risk assessment; route optimization; human travel safety; static and dynamic parameters; risk factor analysis

Paper 41: An Ensemble Machine Learning Model for Predictive Maintenance on Water Injection Pumps in the Oil and Gas Industry

Abstract: The effective operation of water injection pumps is vital for enhancing oil recovery in the oil and gas industry. To ensure optimal pump performance and prevent unplanned downtime, this study focused on implementing predictive maintenance strategies. We began by identifying five critical operational parameters—Seal Pressure 1, Seal Pressure 2, Vibration Data for the Drive End (VIB DE), Vibration Data for the Non-Drive End (VIB NDE), and Ampere. These parameters were monitored and analyzed to evaluate their impact on pump performance and maintenance needs. To achieve this, we applied three machine learning algorithms: Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LGBM), and Random Forest. Each algorithm was independently trained and tested on the dataset corresponding to each operational parameter. We assessed their performance using key accuracy metrics, including R squared, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Following this, we developed an Ensemble model, combining the predictive outputs of XGBoost, LGBM, and Random Forest. The Ensemble model was then applied to the same parameters to evaluate its ability to address the limitations observed in standalone models. The results demonstrated that the Ensemble model consistently delivered superior performance, achieving lower RMSE and MAE values and higher R squared coefficients across all parameters. This study culminates in the validation of the Ensemble model as a robust and reliable approach for predictive maintenance. By leveraging the strengths of multiple algorithms, the Ensemble model offers significant improvements in accuracy and reliability, contributing to more effective maintenance systems for the oil and gas industry.

Author 1: Salama Mohamed Almazrouei

Author 2: Fikri Dweiri

Author 3: Ridvan Aydin

Author 4: Abdalla Alnaqbi

Keywords: Ensemble machine learning models; oil and gas industry; predictive maintenance; water injection pumps

Paper 42: Performance Evaluation of the AuRa Consensus Algorithm for Digital Certificate Processes on the Ethereum Blockchain

Abstract: The blockchain serves as a distributed database where data is stored across different servers and networks. It encompasses various types, with Bitcoin, Ethereum, and Hyperledger being notable examples. To safeguard the security of data transactions on the blockchain, it relies on a consensus algorithm. This algorithm facilitates agreement among nodes within the network. There are multiple types of consensus algorithms, each possessing unique specialties and characteristics. This paper drives into the examination of specific Authority Round, here claimed as AuRa_ori consensus algorithm. The AuRa_ori is a specific type of PoA consensus mechanism used primarily in private or permission blockchain networks. It works by having a set of trusted validators take turns in a round-robin fashion to produce new blocks. It is supported by Parity and Ethereum Clients. AuRa_ori assumes that all the authority nodes are synchronised and honest on every transaction process. In AuRa_ori, every transaction process will execute the four phases i.e., assigning of a new leader, proposing a block, commencing agreement and finally, the phase of committing. However, there exist some discrepancies in some of the phases. In response to the scenario, this paper presents a thorough discussion on the vulnerabilities adhered in AuRa phases in transaction execution by focusing on the first phase of assigning a new leader and the third phase, namely the agreement. The vulnerabilities are subjected to the risk of impacting the performance of Transaction Speed Per Second (TPS), Transaction Throughput (TGS), Percentage Decrease (PD) of TPS and Percentage Increase (PI) of TGS. The new improved method, named AuRa_v1 is parallel presented to overcome the vulnerabilities of AuRa_ori at the selected phases. It aims to increase the TPS and to calculate the PD in transaction process using the Ethereum private blockchain systems. The implementation used three set of data scroll certificate. The result showed that the AuRa_v1 able to decrease the TPS almost 30% based on difference number set of data.

Author 1: Robiah Arifin

Author 2: Wan Aezwani Wan Abu Bakar

Author 3: Mustafa Man

Author 4: Evizal Abdul Kadir

Keywords: Blockchain; Ethereum; consensus algorithm; smart contract; AuRa_ori; AuRa_v1

Paper 43: Enhancing Multiple-Attribute Decision-Making with Interval-Valued Neutrosophic Sets: Diverse Applications in Evaluating English Teaching Quality

Abstract: The evaluation of college English teaching quality is a key method for systematically analyzing and providing feedback on the teaching process and outcomes. It aims to comprehensively assess the effectiveness of teaching, student learning outcomes, and the appropriateness of the course design. The evaluation typically covers aspects such as teaching methods, classroom atmosphere, student engagement, use of teaching resources, and learning achievements. By collecting data from student feedback, teaching supervision, and exam results, the evaluation helps to improve teaching strategies, enhance students' English proficiency, and ultimately achieve continuous optimization and improvement of teaching quality. The teaching quality evaluation of college English is viewed as the multiple-attribute decision-making (MADM). In this paper, some Aczel-Alsina operators are produced under interval-valued neutrosophic sets (IVNSs). Then, interval-valued neutrosophic number (IVNN) Aczel-Alsina weighted averaging (IVNNAAWA) operator is employed to cope with MADM problem. Finally, the numerical decision example for teaching quality evaluation of college English is employed to illustrate the produced method.

Author 1: Lijuan Zhao

Author 2: Shuo Du

Keywords: Multiple-attribute decision-making; interval-valued neutrosophic sets (IVNSs); Aczel-Alsina operations; teaching quality evaluation

Paper 44: A Smoke Source Location Method Based on Deep Learning Smoke Segmentation

Abstract: The generation of smoke is an early warning sign of a fire, and fast, accurate detection of smoke sources is crucial for fire prevention. However, due to the strong diffusivity of smoke, its morphology is easily influenced by environmental factors, and in complex real-world scenarios, smoke sources are often obscured. Current methods lack precision, generalization ability, and robustness in complex environments. With the advancement of deep learning-based smoke segmentation technology, new approaches to smoke source localization have emerged. Smoke segmentation, driven by deep learning models, can accurately capture the morphological characteristics of smoke. This paper proposes a precise and robust smoke source localization method based on deep learning-enabled smoke segmentation. We first conducted experimental evaluations of commonly used deep learning segmentation models and selected the best-performing model as input. Based on the segmentation results, we analyzed the diffusion characteristics and transmittance of smoke, constructed a concentration model, and used it to accurately locate the smoke source. Experimental results demonstrate that, compared with existing methods, this approach maintains high localization accuracy in multi-target smoke scenarios and complex environments, with superior generalization ability and robustness.

Author 1: Yuanpan ZHENG

Author 2: Zeyuan HUANG

Author 3: Hui WANG

Author 4: Binbin CHEN

Author 5: Chao WANG

Author 6: Yu ZHANG

Keywords: Smoke segmentation; smoke source detection; deep learning; instance segmentation; mathematical modeling

Paper 45: Detecting GPS Spoofing Attacks Using Corrected Low-Cost INS Data with an LSTM Network

Abstract: With the emergence of new technologies ranging from smart cities to the Internet of Things (IoT), many objects rely on satellite-based navigation systems, such as GPS, to accomplish their tasks securely. However, GPS receivers are exposed to various unintentional and intentional attacks, threatening the availability and reliability of the delivered information. GPS spoofing is considered as one of the most dangerous attacks, where attackers transmit intense signals on the same frequency to disrupt the GPS receiver, leading to erroneous position calculations. Detection methods for GPS spoofing are crucial to ensure secure navigation. This paper proposes a method for GPS spoofing detection that utilizes artificial intelligence algorithms in combination with raw data from an inertial navigation system (INS). Since INS sensors are prone to accumulating errors over time, these inaccuracies are corrected via a Long Short-Term Memory (LSTM) algorithm. The corrected accelerations and angular rates are then compared to the accelerations and angular rates estimated from the GPS data to detect GPS spoofing signals. This comparison uses the modified M-of-N method, demonstrating its effectiveness by a detection rate reaching 80% of the spoofing zones.

Author 1: Mohammed AFTATAH

Author 2: Khalid ZEBBARA

Keywords: Secure navigation; GPS spoofing; inertial systems; LSTM; M-of-N method; anti-spoofing techniques

Paper 46: Time Distributed MobileNetV2 with Auto-CLAHE for Eye Region Drowsiness Detection in Low Light Conditions

Abstract: Driver drowsiness is a critical factor in road safety, contributing significantly to traffic accidents. This study proposes an innovative approach integrating Auto-CLAHE with Time Distributed MobileNetV2 to enhance drowsiness detection accuracy. This study leveraged the ULg Multimodality Drowsiness Database (DROZY) for facial expression analysis, focusing on the eye region. This study methodology involved segmenting videos into 10-second intervals, extracting 20 images per segment, and applying the Haar Cascade method for eye region detection. The Auto-CLAHE technique was developed to dynamically adjust contrast enhancement parameters based on image characteristics. The analysis yielded promising results. Integrating Auto-CLAHE with Time Distributed MobileNetV2 achieved a classification accuracy of 93.62%, outperforming traditional methods including Greyscale (92.55%), AHE (92.91%), and CLAHE (91.13%). Notably, a precision of 93.71% in detecting drowsiness, with a recall of 93.62% and an F1 score of 93.59% were obtained. Statistical analysis using ANOVA and Tukey HSD tests confirmed the significance of present study results. The key innovation of this study is the implementation of Auto-CLAHE, which significantly improves image contrast adaptation. This approach surpasses AHE and basic CLAHE in drowsiness detection performance, demonstrating remarkable robustness across diverse lighting conditions and facial expressions.

Author 1: Farrikh Alzami

Author 2: Muhammad Naufal

Author 3: Harun Al Azies

Author 4: Sri Winarno

Author 5: Moch Arief Soeleman

Keywords: Driver drowsiness detection; Auto-CLAHE; time distributed; MobileNetV2; eye region analysis

Paper 47: Random Forest Algorithm for HR Data Classification and Performance Analysis in Cloud Environments

Abstract: This study applies the Random forest algorithm to classify and evaluate the effectiveness of business human resources (HR) data, focusing on its potential in supporting strategic decision-making and enhancing organizational efficiency. The research introduces a model that automates the categorization of HR data, including employee records, performance evaluations, and training activities, using the Random Forest method. By constructing both classification and effectiveness assessment models, the study aims to provide businesses with a robust tool for managing and evaluating employee contributions. Key HR metrics were analyzed and categorized, leading to the creation of an effectiveness evaluation model that offers objective insights into employee performance. The Random forest algorithm’s accuracy and stability were validated through cross-validation techniques, proving it to be effective in categorizing employee data and identifying different workforce groups. The models developed in this study are designed to support HR managers in optimizing human resource allocation, improving employee satisfaction, and driving overall business performance. The paper also discusses how the model can be optimized further by expanding data sources and applying it to practical business scenarios.

Author 1: Fangfang Dong

Keywords: Random forest algorithm; business; human resources; data classification

Paper 48: Feature Selection Methods Using RBFNN-Based to Enhance Air Quality Prediction: Insights from Shah Alam

Abstract: This study examines the predictive efficiency of several feature selection approaches in air quality models aimed to predict next-day PM2.5 concentrations in Shah Alam, Malaysia. Air pollution in urban areas is a significant public health concern, and accurate prediction models are essential for timely interventions. However, determining the most important parameters to include in these models remains difficult, especially in complex urban areas with several pollution sources. To address this, we employed three different feature selection methods and applied them to a dataset comprising 43,824 air quality data points provided by the Department of Environmental Malaysia. The data set contained ten variables, such as gas pollutants and meteorological indicators. Each feature selection approach determined top eight variables to include in a Radial Basis Function Neural Network (RBFNN) model. The results showed that ReliefF outperformed Lasso and mRMR in terms of accuracy, specificity, precision, F1 Score, and AUROC, making it the most effective feature selection method for this study. This study contributes to the body of knowledge on air quality modelling by emphasising the relevance of using proper feature selection techniques that are suited to the specific characteristics of the dataset and urban area. Furthermore, it proposes that future study should look into the use of ReliefF-RBFNN in other settings, such as suburban and rural areas, as well as hybrid feature selection approaches to improve prediction performance across several context.

Author 1: Siti Khadijah Arafin

Author 2: Ahmad Zia Ul-Saufie

Author 3: Nor Azura Md Ghani

Author 4: Nurain Ibrahim

Keywords: Lasso; mRMR; PM2.5 concentration; RBFNN; ReliefF

Paper 49: Optimizing CatBoost Model: AI-based Analysis on Rail Transit Figma Platform Practice

Abstract: The research introduces a novel approach that utilizes the Frilled Lizard Optimization (FLO) algorithm to enhance the hyperparameters of the CatBoost model. First, the Figma platform is analyzed in terms of its innovative design applications in rail transit. Then, the FLO algorithm is applied to optimize the CatBoost model, improving its accuracy in detecting foreign objects on rail tracks. Experiments were conducted using a dataset of 6,000 images from rail transit scenarios, divided into seven categories such as left-turning track, straight track, train, pedestrians, and others. The result showed that the FLO-CatBoost model demonstrated superior performance in accuracy, achieving a Root Mean Square Error (RMSE) of 0.274, significantly outperforming other algorithms like TSA, MPA, and RSA. Furthermore, FLO-CatBoost reduced the Mean Absolute Percentage Error (MAPE) and showed better efficiency in evaluation time. Finally, the FLO-CatBoost model significantly enhances the design and evaluation processes for intelligent rail transit systems on the Figma platform, providing higher accuracy and efficiency in detecting foreign objects and improving system design performance.

Author 1: Ruobing Li

Author 2: Hong Qian

Keywords: Rail transport; figma platform innovation design; intelligent analysis and evaluation algorithm; umbrella lizard optimisation algorithm; CatBoost

Paper 50: Color Matching and Light and Shadow Processing in Intelligent Interior Environment Art Design Analysis and Application Based on Neural Network

Abstract: In recent years, the application of Virtual Reality (VR) technology in the field of interior environmental design has expanded significantly, offering designers innovative methods to present complex design concepts within virtual spaces. However, the current color matching and light and shadow processing in reality are not mature enough, and the deep learning algorithms applied in VR are relatively basic with low running efficiency. The consistency and authenticity of virtual reality are not stable enough. This paper explores the integration of color matching and light-shadow processing in interior environmental design within VR technology, with a particular emphasis on leveraging neural network models to achieve automated design optimization. By incorporating deep learning algorithms, this study proposes a neural network-based approach to enhance color matching and light-shadow processing, aiming to improve the realism and aesthetic appeal of virtual environments. Experimental results demonstrate that this method offers substantial advantages in terms of color matching accuracy, naturalness of light-shadow effects, and computational efficiency, highlighting its broad potential for application in virtual reality.

Author 1: Ji Yang

Author 2: Meifen Song

Keywords: Interior environment design; color matching; virtual reality; neural network; light and shadow processing

Paper 51: Selecting the Best Machine Learning Models for Industrial Robotics with Hesitant Bipolar Fuzzy MCDM

Abstract: Machine learning models (MLMs) are used in industry to automate complicated activities, minimize human error, and improve decision-making by evaluating large volumes of data in real time. To managing inventory and quality control in the apparel and auto industries, they provide predictive capabilities such as predicting equipment breakdowns, maintenance and detecting fraud in the finance sector and the major key advantages include cost reduction, higher productivity, better product quality, and tailored client experiences. MLM helps the industries to reduce downtime, prevent errors, and gain a competitive edge through data-driven strategies and processing massive volumes of data in real time. So, there is a need to select the best MLMs for industrial robotics and by considering it, this paper addresses this problem as multiple criteria decision-making (MCDM) by exploiting hesitant bipolar fuzzy information, which takes into account both hesitation and bipolarity in decision-maker preferences. This paper introduced the new aggregation operators (AO) based on geometric and arithmetic procedures to efficiently aggregate the data including the hesitant bipolar fuzzy weighted geometric operator (HBFWGO), which is appropriate for multiplicative relationships, and the hesitant bipolar fuzzy weighted average operator (HBFWAO), which gives weighted importance to qualities. Further, the dual operators including the dual hesitant bipolar fuzzy weighted geometric operator (DHBFWGO) and the dual hesitant bipolar fuzzy weighted average operator (DHBFWAO) have been presented that are further applied to create novel strategies for resolving MCDM issues and offering a methodical manner to assess and combine features. Moreover, the example of selecting the optimal MLMs to show the robustness and efficiency of the suggested methodology has been presented which illustrates the applicability and strength of the proposed methodology in actual decision-making situations.

Author 1: Chan Gu

Author 2: Bo Tang

Keywords: Machine Learning Model (MLM); Hesitant Bipolar Fuzzy Set (HBFS); Dual Hesitant Bipolar Fuzzy Set (DHBFS); Hesitant Bipolar Fuzzy Aggregation Operators (HBFAO); Dual Hesitant Bipolar Fuzzy Aggregation Operators (DHBFAO); Multi-Criteria Decision-Making (MCDM)

Paper 52: Real-Time Data Acquisition in SCADA Systems: A JavaWeb and Swarm Intelligence-Based Optimization Framework

Abstract: This paper aims to improve the accuracy and efficiency of SCADA software design and testing for oil and gas pipelines. It proposes a JavaWeb-based SCADA software solution optimized by the Bird Foraging Search (BFS) algorithm combined with an Echo State Network (ESN) for enhanced testing and analysis. A multi-tiered distributed SCADA software architecture based on the Java EE framework was designed to provide real-time data acquisition, monitoring, control, and data analysis. The BFS algorithm was used to optimize the hyperparameters of the ESN model to improve testing accuracy and convergence speed. The BFS-ESN model was compared with other optimization algorithms such as PSO and DE. Experimental results show that the BFS-ESN model achieved a testing accuracy of 97.33% and faster convergence within 700 iterations. It outperformed other algorithms in both accuracy and convergence speed. The JavaWeb-based SCADA software design for oil and gas pipelines is feasible, and the BFS-ESN model significantly enhances the accuracy and efficiency of SCADA software testing. This approach demonstrates the potential for application in SCADA systems, with future research needed to simplify the model and extend its applicability for large-scale deployment.

Author 1: Lingyi Sun

Author 2: Tieliang Sun

Author 3: Ruojia Xin

Author 4: Feng Yan

Author 5: Yue Li

Author 6: Hengyu Wang

Author 7: Yecen Tian

Author 8: Dongqing You

Author 9: Yun Liu

Author 10: Muhao Lv

Keywords: JAVAWeb; oil and gas pipelines; SCADA software; design analysis; bird foraging search algorithm

Paper 53: Cyber Resilience Model Based on a Self Supervised Anomaly Detection Approach

Abstract: Cyber resilience plays an important role in dealing with cybersecurity and business continuity uncertainty in the post-COVID-19 era. The fundamental problem of cyber resilience is the complexity of real-world problems. Therefore, it is necessary to reduce the complexity of real-world problems to be simple and easy to analyze through cyber resilience model. The first part is the representational model by utilizes world models. It utilizes the stochastic nature of latent data to generate log-likelihood values by data-generating process. The second part is the inference model. This concludes the observation of log-likelihoods using a self-supervised anomaly detection approach. This is related to optimizing decision boundary in anomaly detection, which is achieved by supervising two competing hypotheses based on bias-variance alignment and likelihood ratios. The optimization operates a dynamic threshold supervised by a supervisory signal from the underlying structure of log-likelihoods. The paper contributes by conducting research on the cyber resilience model from the perspective of statistical machine learning. It enhances the representational modeling of world models with the Gaussian mixture model for multimodal regression (GMMR). Additionally, it examines the issue of misleading log-likelihood for out-of-distribution inputs caused by the generalization error and optimizes decision boundary in minimizing the generalization error with a new metric named the harmonic likelihood ratio (HLR). Finally, it aims to boost the performance of anomaly detection using self-supervised learning.

Author 1: Eko Budi Cahyono

Author 2: Suriani Binti Mohd Sam

Author 3: Noor Hafizah Binti Hassan

Author 4: Amrul Faruq

Keywords: Cybersecurity; anomaly detection; cyber resilience model; statistical machine learning; data generating process; bias variance alignment; likelihood ratios; self-supervised learning

Paper 54: Evaluation of the Optimal Features and Machine Learning Algorithms for Energy Yield Forecasting of a Rural Rooftop PV Installation

Abstract: The stability and reliability of the electric grid strongly depend on the ability to schedule and forecast the energy output of all sources. Even though the share of photovoltaic installation in the energy mix is continuously increasing, they have one major drawback: their dependence on different environmental parameters, such as solar irradiance, ambient temperature, cloudiness, etc., which have a highly variable nature. Six machine learning algorithms are compared in this study, regarding their ability to forecast the power generation of a rural rooftop photovoltaic installation using different combinations of the input data. The features selected for investigation are solar radiation, ambient temperature, and wind speed, obtained from a meteorological station, as well as two additional time-based variables – the time of the day and the month of the year. During the validation and testing phases, four models performed better – artificial neural network (ANN), k-Nearest neighbor (kNN), Decision tree (DT), and Random Forest (RF), with ANN achieving the best results in all cases. The optimal combination of input data includes solar radiation, ambient temperature, wind speed, and hour of the day, though the difference with the other scenarios was small. The optimal ANN model achieved R2, MAE, and RMSE of 0.995, 6.71 Wh, and 13.7 Wh, respectively. The results obtained in this study indicate that the yield of PV installations located in rural areas could be forecasted with high probability using a limited number of meteorological data.

Author 1: Boris Evstatiev

Author 2: Katerina Gabrovska-Evstatieva

Author 3: Tsvetelina Kaneva

Author 4: Nikolay Valov

Author 5: Nicolay Mihailov

Keywords: PV yield; forecasting; machine learning; deep learning; features; solar radiation; ambient temperature; wind speed; hour of the day

Paper 55: Edge Computing in Water Management: A KPCA-DeepESN and HOA-Optimized Framework for Urban Resource Allocation

Abstract: This paper presents a novel approach to optimizing urban water resource allocation by integrating Kernel Principal Component Analysis (KPCA) with a Deep Echo State Network (DeepESN), further optimized using the Hiking Optimization Algorithm (HOA). The proposed model addresses the issue of achieving an optimal balance between water supply and demand in urban environments, utilizing advanced machine learning techniques to enhance prediction accuracy and allocation efficiency. KPCA is employed to reduce the dimensionality of key water resource indicators, capturing nonlinear relationships in the dataset. DeepESN, a deep recurrent neural network model, is then applied to predict water consumption trends. HOA, a meta-heuristic algorithm inspired by hiker behavior, is used to fine-tune the DeepESN network parameters, ensuring faster convergence and higher accuracy. The experimental setup includes water resource data from January 2010 to December 2023, divided into training, testing, and validation sets. The model’s performance is compared with other approaches, such as PCA-DeepESN and standalone DeepESN. Results show that the KPCA-HOA-DeepESN model achieves the lowest prediction error and fastest convergence, making it a superior solution for urban water management. Optimized network parameters include a reservoir size of 140, a spectral radius of 0.3, an input scaling factor of 0.22, and a reservoir sparsity degree of 0.72. This study demonstrates the applicability of distributed computing techniques in water resource management by utilizing cloud-based data processing and real-time predictions. The proposed approach not only improves resource allocation but also showcases the potential for edge computing to enhance the responsiveness of water management systems.

Author 1: Hanchao Liao

Author 2: Miyuan Shan

Keywords: KPCA Method; water supply and demand equilibrium; allocation of resources in urban water environment; optimization strategy for hiking; DeepESN

Paper 56: Road Surface Crack Detection Based on Improved YOLOv9 Image Processing

Abstract: Road surface crack detection is a critical task in road maintenance and safety management. Cracks in road surfaces are often the early indicators of larger structural issues, and if not detected and repaired in time, they can lead to more severe deterioration and increased maintenance costs. Effective and timely crack detection is essential to prolong road lifespan and ensure the safety of road users. This paper introduces CrackNet, an advanced crack detection model built upon the YOLOv9 architecture, which integrates a fusion attention module and task space disentanglement to enhance the accuracy and efficiency of road surface crack detection. Traditional methods often struggle with the complex and irregular nature of road cracks, as well as the challenge of distinguishing cracks from their backgrounds. CrackNet overcomes these challenges by leveraging an attention mechanism that highlights relevant features in both the channel and spatial dimensions while separating the tasks of classification and regression. This approach significantly reduces false negatives and improves localization accuracy. The effectiveness of CrackNet is validated through comparative analysis with other segmentation models, including Unet, SOLO v2, Mask R-CNN, and Deeplab v3+. CrackNet consistently outperforms these models in terms of F1 and Jaccard coefficients. This study highlights the critical role of accurate crack detection in minimizing maintenance costs and enhancing road safety.

Author 1: Quanwu Li

Author 2: Shaopeng Duan

Keywords: Road crack; YOLOv9; deep learning; surveillance

Paper 57: DBN-GRU Fusion and Decomposition-Optimisation-Reconstruction Algorithm in Advertising Traffic Prediction

Abstract: As the premise and foundation of advertisement traffic selling and distribution, effective IPTV advertisement traffic prediction not only reduces the operation cost, but also improves the intelligent level of new media advertisement traffic management. In order to further improve the accuracy of new media advertisement traffic prediction, this paper proposes a new media advertisement traffic prediction method based on the hybrid prediction framework of decomposition-optimisation-integration, which is a hybrid model of gated recurrent unit neural network and deep confidence network improved by capsule swarm optimisation algorithm. Firstly, according to the principle of system construction, paper analyses the influencing factors and construct a complete new media advertisement traffic prediction index system; secondly, paper improves the optimisation process of the parameters of the deep confidence network and the gated recurrent unit network by using the quilt group optimisation algorithm, and put forward a new media advertisement traffic prediction method based on the decomposition-optimisation-integration framework; Finally, the proposed method is analysed using new media advertisement traffic data. The results show that the proposed method improves the accuracy of the prediction model and solves the problem of large prediction error in new media advertisement traffic prediction methods.

Author 1: Ronghua Zhang

Keywords: New media advertising traffic prediction; kernel principal component analysis; variational modal decomposition; quilt group algorithm; deep learning; decomposition-optimisation-reconstruction algorithm

Paper 58: Applying Data-Driven APO Algorithms for Formative Assessment in English Language Teaching

Abstract: This study proposes an innovative approach for improving the accuracy and efficiency of formative assessment in English language teaching. The method integrates the Artificial Protozoa Optimization (APO) algorithm with the Kernel Extreme Learning Machine (KELM) to overcome limitations such as local optima in traditional models. The study utilizes data from five university-level English courses, consisting of 327 samples divided into a training set (70%), validation set (15%), and test set (15%). The APO-KELM model is constructed by optimizing the KELM parameters using the APO algorithm. Comparative analysis is conducted against other models, including ELM, KELM, WOA-KELM, PPE-KELM, and AOA-KELM, in terms of accuracy (RMSE), MAPE (Mean Absolute Percentage Error), and convergence speed. The result shows that the APO-KELM model demonstrates superior performance with a Root Mean Square Error (RMSE) of 0.6204, compared to KELM (0.7210), WOA-KELM (0.6934), PPE-KELM (0.6762), and AOA-KELM (0.6451). In terms of MAPE, APO-KELM achieves 0.48, outperforming KELM (0.55), WOA-KELM (0.52), PPE-KELM (0.51), and AOA-KELM (0.49). Additionally, the APO-KELM model converged within 300 iterations, showing faster convergence compared to other models. The integration of the APO algorithm with the KELM significantly enhances the accuracy and efficiency of formative assessment in English language teaching. The APO-KELM model is more accurate and faster than traditional models, making it a valuable tool for improving assessment systems. Future research should focus on refining the APO algorithm for broader applications in educational assessments.

Author 1: Guojun Zhou

Keywords: Big data technology; APO algorithm; formative assessment in English language teaching; nuclear limit learning machine

Paper 59: Enhancing Diabetic Retinopathy Classification Using Geometric Augmentation and MobileNetV2 on Retinal Fundus Images

Abstract: Diabetic retinopathy (DR) ranks among the foremost contributors to blindness worldwide, particularly affecting the adult demographic. Detecting DR at an early stage is crucial for preventing vision loss; however, conventional approaches like fundus examinations are often lengthy and reliant on specialized expertise. Recent developments in machine learning, especially the application of deep learning models, provide a highly effective option for classifying diabetic retinopathy through retinal fundus images. This investigation examines the efficacy of geometric data augmentation methods alongside MobileNetV2 for the classification of diabetic retinopathy. Utilizing augmentation techniques like image resizing, zooming, shearing, and flipping enhances the model's ability to generalize. MobileNetV2 is selected for its impressive inference speed and computational efficiency. This analysis evaluates the effectiveness of MobileNetV2 in relation to InceptionV3, emphasizing metrics such as accuracy, precision, sensitivity, and specificity. The findings show that MobileNetV2 attains exceptional performance, achieving an accuracy of 97%. These findings highlight the promise of employing efficient models and augmentation strategies in clinical settings for the early identification of DR. The findings highlight the critical need to incorporate advanced machine learning methods to enhance healthcare results and avert blindness caused by diabetic retinopathy.

Author 1: Helmi Imaduddin

Author 2: Adnan Faris Naufal

Author 3: Fiddin Yusfida A'la

Author 4: Firmansyah

Keywords: Diabetic retinopathy; data augmentation; InceptionV3; MobileNetV2; transfer learning

Paper 60: New Method in SEM Analysis Using the Apriori Algorithm to Accelerate the Goodness of Fit Model

Abstract: This research aims to develop a new method in Structural Equation Modelling (SEM) analysis using the Apriori algorithm to accelerate the achievement of Goodness of Fit models, focusing on traditional retail purchasing decision models in Indonesia, especially in Palembang. SEM will be used to model causal relationships between variables that influence purchasing decisions in traditional retail. However, the Goodness of Fit model testing process takes a long time due to the complexity of the model. Therefore, this research uses the Apriori algorithm to filter variables that have a significant relationship in traditional retail purchasing decision models to reduce model complexity and speed up Goodness of Fit calculations. There are two stages in the research. First, the Apriori algorithm identifies frequent item sets that frequently appear among variables influencing traditional retail consumer purchasing decisions, such as product, price, and location. This pattern becomes the basis for SEM modeling, focusing on selected variables and, in the second stage, measuring the Goodness of Fit of the SEM model, namely GFI, RMSEA, AGFI, NFI, and CFI, to evaluate the suitability of the model which explains the factors that support traditional retail purchasing decisions in Palembang. The practical implications of this research are significant, as it provides a more efficient and effective method for modeling and understanding consumer behavior in the context of traditional retail. Based on other studies, if this research uses a conventional SEM approach, it does not meet the cut-off value of Goodness of Fit. Meanwhile, the results of the proposed method, namely combining Apriori into SEM, called APR-SEM, obtained a significant Goodness of Fit evaluation. The model coefficient of determination value from APR-SEM is R2 0.71, higher than the conventional model, R2 0.52. This method effectively simplifies the SEM model by identifying the most relevant relationships, thereby providing a clearer understanding of the critical factors influencing purchasing decisions in traditional retail in Palembang City.

Author 1: Dien Novita

Author 2: Ermatita

Author 3: Samsuryadi

Author 4: Dian Palupi Rini

Keywords: APR-SEM; method; goodness of fit; traditional retail

Paper 61: Optimizing Energy Efficient Cloud Architectures for Edge Computing: A Comprehensive Review

Abstract: Now-a-days, edge computing and cloud computing are considered for collaborating together to produce computing solutions that are more effective, scalable and adaptable. The proliferation of cloud infrastructures has drastically increased energy consumption leading to the need for more research in optimizing energy efficiency for sustainable and efficient systems with reduced operational costs. In addition, the edge computing paradigm has gained wide attention during the last few decades due to the rise of the Internet of Things (IoT) devices, the emergence of applications that require low latency, and the widespread demand for environmentally friendly computing. Moreover, lowering cloud-edge systems' energy footprints is essential for fostering sustainability in light of growing concerns about environmental effects. This research presents a comprehensive review of strategies aimed at optimizing energy efficiency in cloud architectures designed for edge computing environments. Various strategies, including workload optimization, resource allocation, virtualization technologies, and adaptive scaling methods, have been identified as techniques that are widely utilized by contemporary research in reducing energy consumption while maintaining high performance. Furthermore, the paper investigates how advancements in machine learning and AI can be leveraged to dynamically manage resource distribution and energy-efficient enhancements in cloud-edge systems. In addition, challenges to the approaches for energy optimization have been discussed in detail to further provide insights for future research. The conducted comprehensive review provides valuable insights for future research in the edge computing paradigm, particularly emphasizing the critical importance of enhancing energy efficiency in these systems.

Author 1: TA Gamage

Author 2: Indika Perera

Keywords: Cloud computing; edge computing; energy efficiency; sustainability

Paper 62: Malicious Traffic Detection Algorithm for the Internet of Things Based on Temporal Spatial Feature Fusion

Abstract: With the rapid development of the Internet of Things, the security issues of its network environment have gradually attracted attention. To enable faster and more accurate identification and detection of malicious traffic attacks in the Internet of Things, an optimized malicious traffic detection algorithm based on fusion of temporal and spatial features is proposed. This method improves the feature extraction performance of traffic data and increases the accuracy of traffic detection. The test results showed that the comprehensive performance of the fusion algorithm was superior to the other four algorithms used for comparison. On the KDD99-CUP dataset, the F1 of the feature fusion algorithm reached 93.16%, while the F1 of algorithms 1-4 were 81.36%, 67.89%, 90.56%, and 92.24%, respectively. On the test set, 182 traffic samples were accurately identified, including 139 correctly identified malicious traffic and 43 correctly identified normal traffic, with recognition accuracy of 98.73% and 97.65%, respectively. Experimental results revealed that the use of fused feature extraction in traffic detection systems could improve detection efficiency and accuracy, providing a safer and more reliable guarantee for the interaction process of the Internet of Things network, and safeguarding the rapid development and application of the Internet of Things.

Author 1: Linzhong Zhang

Keywords: Internet of Things; network security; temporal-spatial characteristics; traffic detection; fusion algorithm

Paper 63: Replace Your Mouse with Your Hand! HandMouse: A Gesture-Based Virtual Mouse System

Abstract: The existing gesture-based operating systems can only simply operate a single piece of software or a specific system, and are not compatible with other applications of mainstream operating systems. In this paper, based on the MediaPipe gesture recognition framework, we design HandMouse, a virtual mouse system that operates using hand gestures. It has the following characteristics: 1. The user does not have contact with the computer hardware when using the system; 2. It requires only one hand to operate, and the design of the gesture considers the ergonomics of the hand; 3. It has most of the functions commonly used in a physical mouse; 4. It can locate the operating area with relative precision. We invited 27 participants to use and evaluate the virtual mouse and then conducted an experiment to compare the performance of the virtual mouse with the physical mouse. The results show that this virtual mouse has a good learning effect and is a great alternative to the physical mouse in public places. The demonstrated operation video is available on https://github.com/wanzhuxie/HandMouse-IJACSA.

Author 1: Qiujiao Wang

Author 2: Zhijie Xie

Keywords: Virtual mouse; ergonomics; gesture; MediaPipe

Paper 64: Deep Learning-Based Network Security Threat Detection and Defense

Abstract: This paper introduces deepnetguard, an innovative deep learning algorithm designed to efficiently identify potential security threats in large-scale network traffic.deepnetguard achieves automated feature learning by fusing basic, statistical, and behavioral features through a multi-level feature extraction strategy, and is capable of identifying both short-time patterns and long-time dependencies. To adapt to the dynamic network environment, the algorithm introduces a dynamic weight adjustment mechanism that allows the model to self-optimize the importance of features based on real-time traffic changes. In addition, deepnetguard integrates auto-encoder (ae) and generative adversarial network (gan) technologies to not only detect known threats, but also recognize unknown threats. By applying the attention mechanism, deepnetguard enhances the interpretability of the model, enabling security experts to track and understand the key factors in the model's decision-making process. Experimental evaluations show that deepnetguard performs well on multiple public datasets, with significant advantages in accuracy, recall, precision, and f1 scores over traditional ids systems and other deep learning models, demonstrating its strong performance in cyber threat detection.

Author 1: Jinjin Chao

Author 2: Tian Xie

Keywords: Network security; threat detection; defense; multilevel feature extraction; dynamic weight adjustment mechanism; interpretability

Paper 65: Development of Fuzzy Logic CRITIC Coupling Coordination Degree Evaluation Algorithm

Abstract: The integrated development of culture and tourism in the Yangtze River Economic Belt refers to a strategic initiative to push economic development and regional coordinated development with culture and tourism as the core. The purpose of this paper is to evaluate the coupling coordination degree of the integrated development of culture and tourism in the Yangtze River Economic Belt, by analysing the integrated development of culture, tourism and economy, and constructing an evaluation index system based on culture and tourism, in which 5 normative indicators and 19 basic indicators are constructed under the cultural perspective, and 4 normative indicators and 10 basic indicators are constructed under the tourism perspective, and its role and impact on regional economic development is explored based on the construction of the index system. Based on the construction of the indicator system, the role and influence of the indicators in regional economic development are explored. The CRITIC algorithm is used to calculate the importance of stratified indicators and stratified evaluation results, and finally, the coupling coordination degree of the research object is calculated through the coupling coordination degree model, which shows that the 13 provinces (municipalities directly under the central government) along the Yangtze River Economic Belt have a slightly different degree of coordination, but the least of them have reached the primary level of coordination, but it also proves that this paper proves the feasibility and necessity of the research method, and it can provide a good solution for the integrated development of culture and tourism in the Yangtze River Economic Belt. However, it also proves the feasibility and necessity of the research method of this paper, which can provide theoretical and practical guidance for the integrated development of culture and tourism in the Yangtze River Economic Belt, and provide new ideas and methods for the development of local tourism along the way.

Author 1: Fangfang Hu

Keywords: Cultural and tourism integration; Yangtze River Economic Belt; coupling coordination degree; CRITIC algorithm

Paper 66: Performance Comparison of Pretrained Deep Learning Models for Landfill Waste Classification

Abstract: The escalating challenge of waste management, particularly in developed nations, necessitates innovative approaches to enhance recycling and sorting efficiency. This study investigates the application of Convolutional Neural Networks (CNNs) for landfill waste classification, addressing the limitations of traditional sorting methods. We conducted a performance comparison of five prevalent CNN models—VGG-16, InceptionResNetV2, DenseNet121, Inception V3, and MobileNetV2—using the newly introduced "RealWaste" dataset, comprising 4,752 labeled images. Our findings reveal that EfficientNet achieved the highest average testing accuracy of 96.31%, significantly outperforming other models. The analysis also highlighted common challenges in accurately distinguishing between metal and plastic waste categories across all models. This research underscores the potential of deep learning techniques in automating waste classification processes, thereby contributing to more effective waste management strategies and promoting environmental sustainability.

Author 1: Hussein Younis

Author 2: Mahmoud Obaid

Keywords: Waste management; deep learning; waste classification; real-waste dataset; performance comparison

Paper 67: Yolov5-Based Attention Mechanism for Gesture Recognition in Complex Environment

Abstract: Object detection is a fundamental task in gesture recognition, involving identifying and localising human hand or body gestures within images or videos amidst varying environmental conditions. To address the inadequate recognition rate of gesture detection algorithms in intricate surroundings caused by issues such as inconsistent illumination, background colors resembling skin tones, and diminutive gesture scales, a gesture recognition approach termed HD-YOLOv5s is presented. An adaptive Gamma image enhancement preprocessing technique grounded in Retinex theory is employed to mitigate the effects of lighting variations on gesture recognition efficacy. A feature extraction network incorporating an adaptive convolutional attention mechanism (SKNet) is developed to augment the network's feature extraction efficacy and mitigate background interference in intricate situations. A novel bidirectional feature pyramid architecture is implemented in the feature fusion network to fully leverage low-level features, thereby minimizing the loss of shallow semantic information and enhancing the detection accuracy of small-scale gestures. A cross-level connection strategy is employed to enhance the model's detection efficiency. To assess the efficacy of the suggested technique, experiments were performed on a custom dataset featuring diverse lighting intensity fluctuations and the publicly available NUS-II dataset with intricate backdrops. The recognition rates attained were 99.5% and 98.9%, respectively, with a detection time per frame of about 0.01 to 0.02 seconds.

Author 1: Deepak Kumar Khare

Author 2: Amit Bhagat

Author 3: R. Vishnu Priya

Keywords: Gesture recognition; Yolov5; object detection; attention mechanism; bidirectional feature pyramid

Paper 68: Multi-Label Aspect-Sentiment Classification on Indonesian Cosmetic Product Reviews with IndoBERT Model

Abstract: For an existing cosmetic company to expand, it is crucial to understand customers’ opinions regarding cosmetic products through product reviews. Aspect-based sentiment classification (ABSC), which consists of text representation and classification stages, is typically employed to automatically extract the interested insights from review. Existing studies of ABSC primarily used single-label classification, which fails to capture relationships between multiple aspects in a review. Additionally, the use of contextual embeddings like IndoBERT for representing Indonesian-language cosmetic product reviews has been underexplored. This study addresses these issues by developing a multi-label classification model that leverages IndoBERT, including IndoBERT[b], IndoBERT[k], and IndoBERTweet, to better represent context and capture relationships across multiple aspects in a review. The model is trained and evaluated using a dataset of Indonesian-language cosmetic product reviews from Female Daily. The multi-label models can be constructed using IndoBERT directly as end-to-end model or employing IndoBERT solely as word embedding model. The latter model, also known as conventional multi-label model, needs to be coupled with problem transformation approach and classifier for classification. Single label classification model with Word2Vec serves as baseline to assess the improvement of multi-label model’s performance on Female Daily cosmetic product reviews dataset. The empirical results revealed that the multi-label approach was more effective in identifying sentiments for pre-defined aspects in reviews. Among the models, end-to-end IndoBERT[b] achieved the highest accuracy (86.98%), while conventional multi-label models combining IndoBERT[b], Label Powerset (LP), and Support Vector Machine (SVM) performed best with 69.64%. This study is significant as it provides a more generalized understanding of the BERT embedding within the context of multi-labels classification and explores the effect of contextual embedding in the cosmetic domain.

Author 1: Ng Chin Mei

Author 2: Sabrina Tiun

Author 3: Gita Sastria

Keywords: Aspect-based sentiment analysis; IndoBERT; multi-label classification; IndoBERTweet; problem transformation

Paper 69: CCNet: CNN CapsNet-Based Hybrid Deep Learning Model for Diagnosing Plant Diseases Using Thermal Images

Abstract: Plant disease diagnosis at an early stage enables farmers, gardeners and agricultural experts to manage and control the spread of illnesses in a timely and suitable manner. The traditional methods of plant disease diagnosis are expensive and might need significant manpower and advanced level machinery. In addition to that, conventional methods, such as visual inspections are prone to subjectivity, time constraints and error susceptibility. In comparison to that, computer based methods such as machine learning is accurately predicting plant diseases underscore the need for a transformative approach. However, by focusing solely on visualized contents and thermal images, these methods overlook the potential insights hidden within customer-posted images that may leads to low accuracy. This study is an attempt to addresses these gaps by proposing an alternative methodology which relies on a hybrid deep learning framework called CCNET. The core CCNET is the utilization of the superiorities of Convolutional Neural capsule network models to get better architecture for plant diseases diagnosis. The proposed CCNET effectively amalgamates the strengths of convolutional layers for spatial feature extraction and the sequential modelling capabilities of CNN and CapsNet for capturing temporal dependencies within image data. The performance of the CCNET has been evaluated through rigorous experimentation. The outcomes underscore the remarkable prowess of the proposed model with the accuracy of 94%. When it compared to the conventional methods, the CCNET surpasses all of them in terms of precision, recall, F-Score, and accuracy.

Author 1: Hassan Al_Sukhni

Author 2: Qusay Bsoul

Author 3: Rami Hasan AL-Taani

Author 4: Fadi yassin Salem Al jawazneh

Author 5: Basma S. Alqadi

Author 6: Misbah Mehmood

Author 7: Asif Nawaz

Author 8: Tariq Ali

Author 9: Diaa Salama AbdElminaam

Keywords: CapsNet; classification; CNN; feature extraction; plant disease; thermal images

Paper 70: A Gradient Technique-Based Adaptive Multi-Agent Cloud-Based Hybrid Optimization Algorithm

Abstract: Efficient virtual machine (VM) movement and task scheduling are crucial for optimal resource utilization and system performance in cloud computing. This paper introduces AMS-DDPG, a novel approach combining Deep Deterministic Policy Gradient (DDPG) with Adaptive Multi-Agent strategies to enhance resource allocation. To further refine AMS-DDPG's performance, we propose ICWRS, which integrates WSO (Workload Sensitivity Optimization) and RSO (Resource Sensitivity Optimization) techniques for parameter fine-tuning. Experimental evaluations demonstrate that ICWRS-enabled AMS-DDPG significantly outperforms traditional methods, achieving a 25% improvement in resource utilization and a 30% reduction in task completion time, thereby enhancing overall system efficiency. By merging nature-inspired optimization techniques with deep reinforcement learning, our research offers innovative solutions to the challenges of cloud resource allocation. Future work will explore additional optimization methods to further advance cloud system performance.

Author 1: Mohammad Nadeem Ahmed

Author 2: Mohammad Rashid Hussain

Author 3: Mohammad Husain

Author 4: Abdulaziz M Alshahrani

Author 5: Imran Mohd Khan

Author 6: Arshad Ali

Keywords: Adaptive multi-agent; cloud-based; hybrid optimization; task scheduling; virtual machine migration; gradient technique

Paper 71: Internet of Things and Cloud Computing-Based Adaptive Content Delivery in E-Learning Platforms

Abstract: In recent years, cloud computing and Internet of Things (IoT) technologies have reshaped e-learning, leading to adaptive content delivery tailored to learners' needs. These paradigms have changed e-learning platforms by providing a scalable and flexible infrastructure for storing and processing large amounts of data. This enables seamless access to teaching materials and resources from anywhere and anytime, increasing the convenience and efficiency of online learning experiences. The convergence of cloud computing, IoT, and e-learning platforms is the heart of this study regarding how these technologies will work together to enable personalized educational experiences. We examine the principles, challenges, and developments in cloud-based adaptive content delivery and highlight the role of IoT data in understanding and incorporating learner preferences. In addition, we discuss possible future directions and implications for the further development of e-learning methods.

Author 1: Lili QIU

Keywords: Cloud computing; Internet of Things; adaptive content delivery; personalized learning; e-learning

Paper 72: Design of a Mobile Language Learning App for Students with ADHD Using Augmented Reality

Abstract: Attention Deficit Hyperactivity Disorder, ADHD for short, impedes submission to traditional teaching as it affects cognitive abilities such as executive function skills, memorization, and focus. In this case, how will kids with ADHD learn languages? This paper provides a solution by presenting the capabilities of a mobile language learning tool called AugmentedFocus, which is designed to support children with ADHD through the use of augmented reality. This association has allowed personalized instruction, accompanied by noticeable augmented reality elements so that learners, teachers, and administrators are able to use their mobile phones to comprehend instructional materials. The objective is to evaluate the application’s design, architecture, prototype, some testing results, and adjustments made during its implementation, while discussing other features within the context. More significantly, we aim to demonstrate how the mobile application can enhance engagement and retention in language learning among kids with ADHD. Attention Deficit Hyperactivity Disorder obstructs dependency on traditional measures of teaching since it relates to cognitive skills like executory function skills, memory, and focus. In this case, how will kids with ADHD learn languages? This paper addresses the issue by demonstrating the effectiveness of a mobile language learning app, AugmentedFocus that is specifically created for ADHD kids and shows a Technology in education.

Author 1: Leonardo Paolo Cesias-Diaz

Author 2: Jorge Armando Laban-Hijar

Author 3: Juan Carlos Morales-Arevalo

Keywords: ADHD; augmented reality; language learning; technology in education; mobile application; design; prototype

Paper 73: Classification of Painting Style Based on Image Feature Extraction

Abstract: The classification of painting style can help viewers find the works they want to appreciate more conveniently, which has a very important role. This paper realized image feature extraction and classification of paintings based on ResNet50. On the basis of ResNet50, squeeze-and-excitation, and convolutional block attention module (CBAM) attention mechanisms were introduced, and different activation functions were selected for improvement. Then, the effect of this method on painting style classification was studied using the Pandora dataset. It was found that ResNet50 obtained the best classification accuracy under a learning rate of 0.0001, a batch size of 32, and 50 iterations. After combining the CBAM attention mechanism, the accuracy rate was 65.64%, which was 6.77% higher than the original ResNet50 and 2.52% higher than ResNet50+SE. Under different activation functions, ResNet50+CBAM (CeLU) had the most excellent performance, with an accuracy rate of 67.13%, and was also superior to the other classification approaches such as Visual Geometry Group (VGG) 16. The findings prove that the proposed approach is applicable to the style classification of painting works and can be applied in practice.

Author 1: Yuting Sun

Keywords: Feature extraction; painting; style classification; ResNet50; attention

Paper 74: Application of Contrast Enhancement Method on Hip X-ray Images as a Media for Detecting Hip Osteoarthritis

Abstract: Image enhancement is one of the most important areas that is being developed in the field of image processing technology. Image contrast enhancement can significantly improve the perception of the digital image itself. X-ray images are crucial in assisting physicians in the formulation of treatment decisions based on diagnostic information. Contrast enhancement techniques, including Histogram Equalization (HE), Contrast Limited Adaptive Histogram Equalization (CLAHE), and CLAHE with double Gamma Correction (CLAGAMTWO), have been utilized on 30 distinct image datasets. Among the three employed methods, the CLAGAMTWO approach yields the optimal values of SSIM = 0.850 and CNR = 0.773. CLAHE has superior performance with an Entropy value of 7.099. CLAGAMTWO is the superior approach overall, as evidenced by the average metric value, yielding optimal picture quality in visual structure (SSIM), information detail (Entropy), and crisp contrast with little noise (CNR).

Author 1: Faisal Muttaqin

Author 2: Jamari

Author 3: R Rizal Isnanto

Author 4: Tri Indah Winarni

Author 5: Athanasius Priharyoto Bayuseno

Keywords: X-ray; image enhancement; digital image; image processing; grayscale image

Paper 75: Computer-Vision-Based Detection and Monitoring System for Mature Coconut Fruits with a Web Dashboard Visualization Platform

Abstract: The Philippines is the second largest producer of coconut products in the world with 347 million trees planted in 3.6 million hectares of land across the country. Traditionally, harvesting coconuts is a labor-intensive process in the Philippines that involves manual climbing and chopping fruits, which carries a high risk of harm or even death. Hence, the number of expert coconut climbers has decreased as a result. In response, current research has concentrated on creating robot harvesters. However, classifying the mature coconut fruit is a major problem in the harvesting process that calls for a great deal of experience, patience, and work. Studies employing Convolutional Neural Networks (CNNs) have shown great accuracy in detecting coconut ripeness, although these efforts have been limited to detection without practical integration with harvesting equipment. Moreover, the present research lacks a comprehensive solution that allows real-time data display and monitoring, such as the maturation stage of coconuts, via a web-based dashboard. This discrepancy emphasizes the requirement for systems that can not only identify the age of coconuts but also work with harvesting technologies and provide intuitive user interfaces for data display and decision-making. In order to fill these gaps, this study presents a computer-vision-based system that monitors and detects coconut fruit maturity, with an emphasis on mature coconuts, by utilizing the YOLOv8 model. With a Mean Average Precision (mAP50) of 99.5%, mAP50-95 of 89.5%, precision of 99.5%, and recall of 99.9%, the system demonstrated great accuracy. A web-based dashboard is also integrated into the system to provide monitoring and visualization of detected coconut fruits, along with notifications for fully ripe fruits.

Author 1: Samfford S. Cabaluna

Author 2: Maria Fe P. Bahinting

Author 3: Leah A. Alindayo

Keywords: Coconut fruit maturity; coconut maturity detection; computer vision; crop monitoring

Paper 76: Simulation Analysis of Intelligent Control System for Excavators in Large Mining Plants Based on Electronic Control Technology

Abstract: With the increasing demand for large-scale mine equipment and the complexity of the operating environment, the intelligent trajectory planning and control of mine systems becomes very important. This paper proposes a proportional-integral-differential (PID) feedback controller combined with adaptive improvement. This controller combines Genetic Algorithm and Particle Swarm Optimization technology to enhance the ability of the excavator’s intelligent control system and improve the control accuracy, response speed, and robustness under different working conditions. The results showed that the constructed PID controller improved the average constraint performance by 2.5% through quintic interpolation, and the power consumption was relatively small. The trajectory prediction error of different joints was less than 5% and the displacement and pressure curves of the hydraulic cylinder were stable and symmetrical. The accuracy of the proposed algorithm was 94% and quickly converged to 0.05 after 50 iterations, which was 18.5%, 15.3%, and 17.5% higher than the other three algorithms, respectively. Therefore, the proposed method has high reliability and adaptability in anti-interference ability, trajectory planning progress, and optimization efficiency, and it provides a better solution for intelligent control of the excavator excavation system.

Author 1: Lei Sun

Keywords: Genetic Algorithm; Particle Swarm Optimization; proportional-integral-differential; mining system; intelligent control

Paper 77: Predicting Graft Failure Within Year After Transplantation Using Data Mining Techniques

Abstract: The complex factors of liver transplant survival and the potential for post-transplant complications are significant challenges for healthcare professionals. This paper aims to identify the ability to use data mining techniques to develop a predictive model for liver transplant failure by identifying the relationship between abnormalities in periodic patients' laboratory results and graft failure. The researchers obtained data from King Faisal Specialist Hospital and Research Centre to address the research problems. First, the classification technique was used to predict cases with a high risk of liver transplant failure. Second, Association Rules were applied to identify associations between abnormalities in patients’ laboratory results and transplant failure. Before using data mining algorithms, the patient dataset underwent a cleaning process, which involved removing duplicate entries and uncertain results. The algorithms were applied separately to the data of patients who completed the first year without complications and those who experienced transplant failure. The obtained results were then compared and we observed that abnormal levels in Aspartate Transferase (AST), Red Blood Cell (RBC), Hemoglobin (Hgb), 'Bilirubin Total', and 'Platelet' occurred exclusively in cases that faced liver transplant failure within the first year. Similarly, abnormal levels in 'AST', 'RBC', Alanine Aminotransferase (ALT), and 'Bilirubin Total' were also associated with transplant failure.

Author 1: Meshari Alwazae

Author 2: Saad Alghamdi

Author 3: Lulu Alobaid

Author 4: Bader Aljaber

Author 5: Reem Altwaim

Keywords: Graft failure; liver transplant; data mining; predictive model; classification; association rules

Paper 78: Synthesizing Realistic Knee MRI Images: A VAE-GAN Approach for Enhanced Medical Data Augmentation

Abstract: This study presents a novel approach for synthesizing knee MRI images by combining Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs). By leveraging the strengths of VAEs for efficient latent space representation and GANs for their advanced image generation capabilities, we introduce a VAE-GAN hybrid model tailored specifically for medical imaging applications. This technique not only improves the realism of synthesized knee MRI images but also enriches training datasets, ultimately improving the outcome of machine learning models. We demonstrate significant improvements in synthetic image quality through a carefully designed architecture, which includes custom loss functions that strike a balance between reconstruction accuracy and generative quality. These improvements are validated using quantitative metrics, achieving a Mean Squared Error (MSE) of 0.0914 and a Fréchet Inception Distance (FID) of 1.4873. This work lays the groundwork for novel research guidelines in biomedical image study, providing a scalable solution to overcome dataset limitations while maintaining privacy standards, and pavement of reliable diagnostic tools.

Author 1: Revathi S A

Author 2: B Sathish Babu

Keywords: Custom loss function; decoder; discriminator; GAN; latent space; VAE

Paper 79: Enhancing Mobility – An Intelligent Robot for the Visually Impaired

Abstract: Efficient robot navigation in operational environments requires precise tracking of the path from the starting point to the destination, typically generated using pre-stored map data. However, obstacles in the environment can complicate this process, making reliable obstacle avoidance critical for successful navigation. This paper introduces innovative techniques for robotic navigation and obstacle avoidance, specifically designed to assist visually impaired individuals. To mitigate the limitations and inaccuracies inherent in sensor data, we employ sensor fusion algorithms that integrate inputs from infrared, ultrasonic, vision, and tactile sensors. Additionally, visual landmarks are incorporated as reference points to improve internal odometry correction and enhance mapping accuracy. We believe that our approach not only increases the reliability of navigation but also enhances the robot's ability to operate effectively in diverse and challenging conditions.

Author 1: Ahmad M. Bisher

Author 2: Rufaida M. Shamroukh

Author 3: Abed M. Shamroukh

Keywords: Robot; obstacle; avoidance; visually impaired; sensors

Paper 80: Face Anti-Spoofing Using Chainlets and Deep Learning

Abstract: Now-a-days, biometric technology is widely employed for many security purposes. Facial recognition is one of the biometric technologies that is increasingly utilized because it is convenient and contactless. However, the facial recognition system has become the most targeted by unauthorized users to get access to the system. Most facial recognition systems are vulnerable to face spoofing attacks. With the widespread use of the internet and social media, it has become easy to get videos or pictures of people’s faces. The imposter can use these documents to deceive facial authentication systems, which affects the system’s security and privacy. Face spoofing occurs when an unauthorized user attempts to gain access to a facial recognition system using presentation attack instruments (PAIs) such as photos, videos, or 3D masks of the authorized users. Therefore, the need for an effective face anti-spoofing (FAS) system is increased. That motivated us to develop a face anti-spoofing model that accurately detects presentation attacks. In our work, we developed a model that integrates handcrafted features based on Chainlets (as motion-based descriptor) and the convolutional neural network (CNN) to provide a more robust feature vector and enhance accuracy performance. Chainlets can be computed from deep contour-based edge detection using Histograms of Freeman Chain Codes, which provides a richer and rotation-invariant description of edge orientation that can be used to extract Chainlets features. We used a benchmark dataset, the Replay-Attack database. The result shows that the Chainlets-based face anti-spoofing method overcome the state-of-art methods and provide higher accuracy.

Author 1: Sarah Abdulaziz Alrethea

Author 2: Adil Ahmad

Keywords: Presentation attacks; Chainlets; contour; handcrafted features; chain code; CNN; face anti-spoofing

Paper 81: DSTC-Sum: A Supervised Video Summarization Model Using Depthwise Separable Temporal Convolutional

Abstract: The exponential growth in video content has created a critical need for efficient video summarization techniques to enable faster and more accurate information retrieval. Video summarization has excellent potential to simplify the analysis of large video databases in various application areas ranging from surveillance, education, entertainment, and research. DSTC-Sum, a novel supervised video summarization model, is proposed based on Depthwise Separable Temporal Convolutional (DSTC). Leveraging the superior representational efficiency of DSTCN, the model addresses computational challenges and training inefficiencies encountered in traditional recurrent architectures such as Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTMs). Additionally, this approach reduces computational overhead and memory usage. DSTC-Sum achieved state-of-the-art performance on two commonly used benchmark datasets, TVSum and SumMe, and outperformed all previous methods with F-scores by 1.8% and 3.33%, respectively. To validate the model's generality and robustness, the model was further tested on the YouTube and Open Video Project (OVP) datasets. The proposed model did slightly better on these datasets than several popular techniques, with F scores of 60.3 and 58.5, respectively. Finally, these findings confirm that this model captures long-term temporal dependencies and produces high-quality video summaries across all types of videos.

Author 1: M. Hamza Eissa

Author 2: Hesham Farouk

Author 3: Kamal Eldahshan

Author 4: Amr Abozeid

Keywords: Video summarization; depthwise separable temporal convolutional; video processing; deep learning

Paper 82: A Taxonomic Study: Data Placement Strategies in Cloud Replication Environments

Abstract: Since the past decades, the data replication trend has not subsided; it is progressing rapidly from multiple perspectives to enhance cloud replication performance. Researchers are eagerly focusing on improving the strategies in various perceptions; unfortunately, the vulnerability in every strategy is inevitable. A non-comprehensive replica strategy would have vulnerability and drawbacks. The drawbacks that usually reside in the developed strategies are not limited to high network usage, high process time, high response time, high storage consumption, and more, depending on the research areas. Many researchers are out of ideas to identify state-of-the-art issues. This exhaustive taxonomic study focused on analyzing the diversified contributions and limitations terrain of the cloud replication environment, focusing on data placement strategies. It seeks to delve deeply into its fundamental strategy, practical implementations, and the intricate challenges it poses. Concerning the imminent cloud-driven future, this structured review paper is a vital resource for researchers, policymakers, and industry professionals grappling with the complexities of this emerging paradigm. By illuminating the intricacies of data replication strategies, this study fosters a deeper appreciation for the transformative potential and the multifaceted challenges ahead of cloud data replications.

Author 1: Fazlina Mohd Ali

Author 2: Marizuana Mat Daud

Author 3: Fadilla Atyka Nor Rashid

Author 4: Nazhatul Hafizah Kamarudin

Author 5: Syahanim Mohd Salleh

Author 6: Nur Arzilawati Md Yunus

Keywords: Cloud environment; data replication; placement strategies; replication taxonomy; performance metrics

Paper 83: Optimizing House Renovation Projects Using Industrial Engineering-Based Approaches

Abstract: The persistent challenge of project delays poses significant issues with the escalating demand for house renovations. The company in Kedah, Malaysia, faces frequent project delays due to ineffective project management, leading to substantial liquidated damages. This study aims to minimise project delays and ensure timely completion within budget constraints, focusing on both the entire house renovation project and the kitchen renovation project. This study employed the Program Evaluation and Review Technique to illustrate the project network and utilised the Critical Path Method to identify the critical path while implementing project crashing with linear programming to optimise activity duration reduction and minimise costs. The PERT method results in an illustrative network diagram that aids subsequent analysis. The completion time for the entire house renovation project is determined to be 58 days using CPM, with a 96.8% probability of completion within 60 days. In contrast, for the kitchen renovation project, the completion time is identified as 38 days, with a 0% probability of meeting the 30-day deadline. Therefore, linear programming was successfully applied, shortening the kitchen project to 30 days at a total cost of RM 18,517.50, further reduced to 20 days with a cost of RM 20,980. Both scenarios remained below the total penalty cost of RM 21,780. The finding enables the company to make informed decisions on resource allocation to accelerate project duration and avoid delays. Future research should delve into realistic models, considering labour allocation and indirect costs, for a more comprehensive evaluation of project crashing strategies and their financial impacts.

Author 1: Lim Rou Yan

Author 2: Siti Noor Asyikin Mohd Razali

Author 3: Muhammad Ammar Shafi

Author 4: Norazman Arbin

Keywords: House renovation; Program Evaluation and Review Technique (PERT); Critical Path Method (CPM); project crashing techniques; construction project management; linear programming optimisation

Paper 84: FSFYOLO: A Lightweight Model for Forest Smoke and Fire Detection

Abstract: The detection and identification of forest smoke and fire are critical for forest fire prevention efforts. However, current forest smoke and fire target detection algorithms confront obstacles such as high memory usage, computational costs, and deployment difficulty. Regarding these key issues, this paper presents FSFYOLO, a lightweight forest smoke and fire detection model based on the YOLOv8s model. To efficiently extract key features from forest smoke and fire images while reducing computational redundancy, the lightweight network EfficientViT is used as the backbone network. A lightweight detection head, Partial Convolutional Head (PCHead), is designed using the shared parameters idea to greatly minimize the amount of parameters and computations by leveraging shared convolutional layers and branched processing, thus achieving the lightweight design of the model. In the neck network, a lightweight feature extraction module, C2f-FL, is built to more fully extract local features and surrounding contextual information to widen the receptive field. Additionally, a Coordinate Attention (CA) mechanism is integrated into both the backbone and neck networks to capture cross-channel information, directional awareness, as well as position-sensitive information, improving the model's capacity to precisely pinpoint fire and smoke in forests. The experimental outcomes results on our self-constructed forest smoke and fire dataset demonstrate that FSFYOLO reduces the number of parameters and computation by 47.6% and 60.9%, respectively, compared to the original model, while improving precision, recall, and mAP50 by 1.3%, 1.0%, and 1.0%, respectively. This demonstrates that FSFYOLO strikes a good compromise between model lightweighting and detection accuracy.

Author 1: Yinglai HUANG

Author 2: Jing LIU

Author 3: Liusong YANG

Keywords: Forest smoke and fire; target detection; lightweight; YOLOv8; EfficientViT

Paper 85: Identification of Chili Plant Diseases Based on Leaves Using Hyperparameter Optimization Architecture Convolutional Neural Network

Abstract: This paper proposes a method to detect chili plant diseases based on leaves. Studies in recent years have shown that chili production in Indonesia has decreased. This is because there are several influencing factors. One common factor is the presence of diseases in chili plants that cause less than optimal harvest production. Fungi or pests on chili leaves usually cause diseases that often appear in chili plants. Chili leaf diseases have a negative impact on chili harvest yields. Chili leaf diseases can result in significant decreases in both the quantity and quality of chili harvests. Accurate disease diagnosis will help increase farmer profits. This study identified four major leaf diseases, namely leaf curl, leaf spot, yellowish, and white spot. In this research images were taken using a digital camera. These diseases were classified into five classes (healthy, leaf curl, leaf spot, yellowish, and white spot) using two different pre-trained deep learning networks, namely MobileNetV2 and VGG16, using chili leaf data through deep learning transfer. The experimental results showed the model with the best performance was the VGG16 model. This model achieved a validation accuracy of 94% on public and own data sets. Meanwhile, the next best-performing model is MobileNetV2, which achieved an accuracy of 90%, followed by the Traditional CNN Model, which achieved a validation accuracy of 88%. In future developments, we intend to deploy it on mobile devices to automatically monitor and identify various types of chili plant disease information based on leaves.

Author 1: Murinto

Author 2: Sri Winiarti

Author 3: Ardi Pujiyanta

Keywords: Chili leaf; deep learning; MobileNetV2; transfer learning; VGG16

Paper 86: The Application of K-MEANS Algorithm-Based Data Mining in Optimizing Marketing Strategies of Tobacco Companies

Abstract: With the continuous development of data mining technology, more and more industries are applying data mining techniques to optimize their marketing strategies. In response to the persistent decline in tobacco sales and the gradual erosion of customer base in a particular enterprise in recent years, this study employs data mining technology to enhance the current tobacco marketing strategy. Firstly, in response to the current shortcomings of the company, a marketing optimization design scheme was proposed and a customer classification evaluation index system was constructed. Subsequently, homomorphic encryption technology and enhanced peak density thinking were employed to enhance the conventional K-means algorithm. The enhanced algorithm was then utilized in the customer clustering and partitioning scheme, with the objective of investigating the underlying information present in customer consumption data. The performance of the algorithm was tested, and the results showed that the mean square error of the improved K-means algorithm was about 0.1, with an average absolute error of about 0.05. The highest detection rate in the validation set was 0.95, and the lowest false alarm rate was 0.07. Both experts and customers were highly satisfied with the marketing strategy under the enhanced K-means algorithm. In summary, the clustering analysis method used in this study can effectively uncover the hidden value behind various types of customer data, thereby helping companies to make better marketing strategies.

Author 1: Mingqian Ma

Keywords: Data mining; homomorphic encryption; k-means; tobacco; marketing strategy; indicator system

Paper 87: Application of Machine Learning Algorithms for Predicting Energy Consumption of Servers

Abstract: Energy management in data centers is currently a major challenge and arouses considerable interest. Many data center operators are seeking solutions to reduce energy consumption. In this work, the problem of resource overutilization-defined as the excessive usage of critical server resources such as CPU, RAM and storage surpassing their optimal capacity-in data centers is addressed, with a particular focus on servers. Estimating the energy consumption of servers in data centers allows its managers to allocate the necessary resources to ensure adequate quality of service. The research involved generating workloads performance on various servers, each connected to a wattmeter for energy consumption measurement. Data on resource utilization rates and server energy consumption were stored and analyzed. Machine learning models were then used to forecast server energy consumption. Parametric, non-parametric, and ensemble methods were employed and validated using accuracy measurements, non-parametric tests, and model complexity to assess the quality of energy consumption prediction models. The results demonstrated that certain models could provide predictions with a low margin of error and minimal complexity like polynomial regression, while other models showed lower performance. A comparative analysis is conducted to evaluate the performance and limitations of each approach.

Author 1: Meryeme EL YADARI

Author 2: Saloua EL MOTAKI

Author 3: Ali YAHYAOUY

Author 4: Khalid EL FAZAZY

Author 5: Hamid GUALOUS

Author 6: Stéphane LE MASSON

Keywords: Data center; server; machine learning; energy consumption; parametric methods; ensemble methods

Paper 88: CQRS and Blockchain with Zero-Knowledge Proofs for Secure Multi-Agent Decision-Making

Abstract: Autonomous decision-making in decentralized multi-agent systems (MAS) poses significant challenges related to security, scalability, and privacy. This paper introduces an innovative architecture that integrates Decentralized Identifiers (DIDs), Zero-Knowledge Proofs (ZKPs), Hyperledger Fabric blockchain, OAuth 2.0 authorization, and the Command Query Responsibility Segregation (CQRS) pattern to establish a secure, scalable, and privacy-focused framework for MAS. The use of DIDs and ZKPs ensures secure, self-sovereign identities and enables privacy-preserving interactions among autonomous agents. Hyperledger Fabric provides an immutable ledger, ensuring data integrity and facilitating transparent transaction processing through smart contracts. The CQRS pattern, combined with event sourcing, optimizes the system’s ability to handle high volumes of read and write operations, enhancing performance and scalability. Practical applications are showcased in Smart Grids, Healthcare Data Management, Secure Internet of Things (IoT) Networks, and Supply Chain Management, highlighting the architecture’s ability to address industry-specific challenges. This integration offers a robust solution for ensuring trust, verifiability, and scalability in distributed systems while preserving the confidentiality of agents.

Author 1: Ayman NAIT CHERIF

Author 2: Mohamed YOUSSFI

Author 3: Zakariae EN-NAIMANI

Author 4: Ahmed TADLAOUI

Author 5: Maha SOULAMI

Author 6: Omar BOUATTANE

Keywords: Decentralized multi-agent systems; decentralized identifiers; zero-knowledge proofs; hyperledger fabric; OAuth 2.0; CQRS; smart grids; healthcare data management; IoT; supply chain management

Paper 89: An Efficient Privacy-Preserving Randomization-Based Approach for Classification Upon Encrypted Data in Outsourced Semi-Honest Environment

Abstract: In cloud environment context, organizations often rely on the platform for data storage and on demand access. Data is typically encrypted either by the cloud service itself or by the data owners before outsourcing it to maintain confidentiality. However, when it comes to processing encrypted data for tasks like kNN classification; existing approaches either prove to be inefficient or delegate portion of the classification task to end users or do not satisfy all the privacy requirements. Also, the datasets used in many existing approaches to check the performance seem to have very less attributes and instances; but, it is observed that as dataset size increases, the efficiency and accuracy of many privacy-preserving approaches reduce significantly. In this work, we propose a set of privacy preserving protocols that collectively perform the kNN classification with encrypted data in outsourced semi-honest-cloud environment and also address the stated challenges. This is accomplished by building an efficient randomization-based approach called PPkC that leverages homomorphic cryptosystem properties. With protocol analysis we prove that the proposed approach satisfies all privacy requirements. Finally, with extensive experimentation using real-world and scaled dataset we show that the performance of proposed PPkC protocol is computationally efficient and also independent of the number of nearest neighbours considered.

Author 1: Vijayendra Sanjay Gaikwad

Author 2: Kishor H. Walse

Author 3: Mohammad Atique Mohammad Junaid

Keywords: Partial homomorphic encryption; classification using encrypted data; randomization; k- nearest neighbours

Paper 90: Modeling the Impact of Robotics Learning Experience on Programming Interest Using the Structured Equation Modeling Approach

Abstract: Proficiency in programming is crucial for driving the Fourth Industrial Revolution. Therefore, interest in programming needs to be instilled in students starting from the school level. While the use of robotics can attract students' interest in programming, there is still a lack of research modeling, the impact of robotic learning experiences on programming interest using a structural equation modeling (SEM) approach. This study aims to analyze the structural relationship between interest in programming and learning experiences using a specially developed robotics module based on Kolb's experiential learning model and the programming development phases. An experiment involving 76 primary and secondary school students was conducted using the robotics module. Data were collected through a questionnaire containing 12 questions for five constructs: engagement, interaction, challenge, competency, and interest. These constructs, which are latent variables, formed the model using the partial least squares-SEM technique through the SmartPLS 4.0 software. The evaluation of the structural model found that the variables of engagement and competency had a significant impact on interest in programming, while interaction and challenge received low values. The developed model has moderate predictive power, indicating that interest in programming can be moderately predicted based on students' experiences using robots.

Author 1: Nazatul Aini Abd Majid

Author 2: Noor Faridatul Ainun Zainal

Author 3: Zarina Shukur

Author 4: Mohammad Faidzul Nasrudin

Author 5: Nasharuddin Zainal

Keywords: Programming; robotics; Structural Equation Modeling (SEM); experiential learning; student engagement

Paper 91: Lampung Batik Classification Using AlexNet, EfficientNet, LeNet and MobileNet Architecture

Abstract: This study explores the application of image recognition technology based on Convolutional Neural Network (CNN) to classify Lampung batik motifs. Four CNN architectures are employed, namely AlexNet, EfficientNet, LeNet, and MobileNet. The dataset consist of ten motif classes, including Siger Ratu Agung, Sembagi, Jung Agung, Kembang Cengkih, Granitan, Abstract, Sinaran, Tambal, Kambil Sicukil, and Sekar Jagat. It comprises a total of 1000 images of Lampung Batik motifs, which were enhanced using preprocessing techniques such as rotation, shifting, brightness adjustment, and zooming. The classification results show that AlexNet achieves an accuracy of 95.33%, EfficientNet achieves 98.00%, LeNet achieves 99.33%, and MobileNet achieves 98.00%. The best accuracy result was achieved by the LeNet architecture, attributed to its suitability for small datasets. While some classification errors occurred due to similarities in patterns and variations in image positions, employing more advanced methods to better distinguish between similar motifs could address these challenges. This study highlights the effectiveness of CNN architectures in supporting the recognition of Lampung Batik motifs, contributing to the understanding and preservation of Indonesia's cultural heritage.

Author 1: Rico Andrian

Author 2: Rahman Taufik

Author 3: Didik Kurniawan

Author 4: Abbie Syeh Nahri

Author 5: Hans Christian Herwanto

Keywords: Lampung Batik; image classification; convolutionl neural network; AlexNet; EfficientNet; LeNet; MobileNet

Paper 92: Optimization of DL Technology for Auxiliary Painting System Construction Based on FST Algorithm

Abstract: The continuous development of computers has brought about the emergence of many image processing software, but these software have relatively limited functions and cannot learn and create works according to the prescribed style. To make it easier for ordinary people to create artistic style paintings, this study proposes the construction of an auxiliary painting system based on finite state transducer algorithm-optimized deep learning technology. The results demonstrated that when there were 12 images, the accuracy of the optimized convolutional neural network model in extracting image features increased by 1.1% compared to before optimization. When the number of images was 1, the optimized model reduced the image feature extraction time by 15.1s compared to before optimization. Compared with other algorithms, the accuracy of extracting image style information based on a convolutional neural network was the highest at 80% under different iteration times. The research algorithm has improved the accuracy and time of extracting image style information.

Author 1: Pengpeng Xu

Author 2: Guo Chen

Keywords: Finite state transducer; deep learning; CNN; auxiliary painting; style transfer

Paper 93: BackC&P: Augmenting Copy and Paste Operations on Mobile Touch Devices Through Back-of-device Interaction

Abstract: As more and more complex applications, e.g. photo editing software and slideshow editing software, can be used on mobile touch devices, some simple operations, such as copying and pasting, are used more frequently by ordinary mobile users. However, the existing touch techniques are far from perfectly supporting these simple operations on mobile devices. In this paper, a new interactive technique BackC&P, which takes advantage of back-of-device touch input to augment copy and paste operations on mobile devices, is presented. The results of a user study that evaluated the usability of BackC&P are also presented. The findings indicate that BackC&P was about twice as fast as the currently used technique on mobile touch devices when used to complete the copy-and-paste tasks, with no significant decrease in accuracy.

Author 1: Liang Chen

Keywords: Back-of-device interaction; copy and paste operations; mobile touch devices; touch interaction

Paper 94: A Review: PTSD in Pre-Existing Medical Condition on Social Media

Abstract: Post-Traumatic Stress Disorder (PTSD) is a multifaceted mental health condition, particularly challenging for individuals with pre-existing medical conditions. This review critically examines the intersection of PTSD and chronic illnesses as expressed on social media platforms. By systematically analyzing literature from 2008 to 2024, the study explores how PTSD manifests and is managed in individuals with chronic conditions such as cancer, heart disease, and autoimmune disorders, with a focus on online expressions on platforms like X (formally known as Twitter) and Facebook. Findings demonstrate that social media data offers valuable insights into the unique challenges faced by individuals with both PTSD and chronic illnesses. Specifically, natural language processing (NLP) and machine learning (ML) techniques can identify potential PTSD cases among these populations, achieving accuracy rates between 74% and 90%. Furthermore, the role of online support communities in shaping coping strategies and facilitating early interventions is highlighted. This review underscores the necessity of incorporating considerations of pre-existing medical conditions in PTSD research and treatment, emphasizing social media's potential as a monitoring and support tool for vulnerable groups. Future research directions and clinical implications are also discussed, with an emphasis on developing targeted interventions.

Author 1: Zaber Al Hassan Ayon

Author 2: Nur Hafieza Ismail

Author 3: Nur Shazwani Kamarudin

Keywords: PTSD; mental health; social media; natural language processing; health informatics

Paper 95: CIPHomeCare: A Machine Learning-Based System for Monitoring and Alerting Caregivers of Cognitive Insensitivity to Pain (CIP) Patients

Abstract: Congenital Insensitivity to Pain (CIP) patients, particularly infants, are vulnerable to self-injury due to their inability to perceive pain, which can lead to severe harm, such as biting their hands. This research introduces "CIPHomeCare," a wearable monitoring solution designed to prevent self-injurious behaviors in CIP patients aged 6 to 24 months. The primary focus of this study is developing and applying machine learning algorithms to classify hand-biting behaviors. Using accelerometer data from the STEVAL-BCN002V1 sensor, which is a motion sensor, several machine learning models—K-Nearest Neighbors (KNN), Random Forest (RF), Naive Bayes (NB), Linear Discriminant Analysis (LDA), and Logistic Regression (LR)—were trained to differentiate between normal and harmful behaviors. To address data imbalance due to the infrequency of biting events, oversampling techniques such as SMOTE, Borderline-SMOTE, ADASYN, K-means-SMOTE, and SMOTE-ENN were employed to enhance classification performance. Among the algorithms, KNN achieved the highest accuracy (98%) and a sensitivity of 72%, highlighting its effectiveness in detecting harmful hand motions. The findings suggest that machine learning, in combination with wearable technology, can provide accurate, personalized monitoring and timely intervention for CIP patients, paving the way for broader clinical applications and real-time prevention of self-injury. The real-time processing capability of the system enables immediate alerting of caregivers, allowing for timely intervention to prevent injuries, thus improving their quality of life.

Author 1: Rahaf Alsulami

Author 2: Hind Bitar

Author 3: Abeer Hakeem

Author 4: Reem Alyoubi

Keywords: Cognitive insensitivity to pain patients; CIP; machine learning; motion sensors; quality of life; wearable activity recognition

Paper 96: A Multi-Person Collaborative Design Method Driven by Augmented Reality

Abstract: The current interior design of commercial buildings is facing innovative challenges, requiring a balance between aesthetics, functionality, and economic benefits. The design industry faces challenges in interdisciplinary integration, lack of standardized processes, and limitations of traditional design methods in complex situations. Although virtual reality technology provides new solutions, its integration difficulty, cost, and operational complexity constrain its widespread application. This study introduces a digitally twin-based augmented reality (AR) collaborative indoor design framework, addressing the decomposition of spatial planning for the complexity inherent in design processes. Subsequently, contextual data in the indoor design process is structured into an indoor design knowledge graph to elucidate information transmission and iterative mechanisms during collaborative design, thereby enhancing the situational adaptability of AR collaborative design. Utilizing a root anchor-based collaborative approach, multiple designers engage in spatial design collaboration within the AR environment. Real-time knowledge and data facilitated by the design knowledge graph contribute to collaborative decision-making, ensuring the quality and efficiency of collaborative design. Finally, exemplified by a complex interior design project for a commercial space, an AR-based collaborative digital twin (DT) interior design system is established, validating the effectiveness and feasibility of the proposed methodology. Through this approach, designers can preview and modify designs in a virtual environment, ultimately reducing errors, shortening design cycles, lowering costs, and enhancing user satisfaction.

Author 1: Liqun Gao

Keywords: Digital twin; optimized design; interior space; multi-person collaborative design

Paper 97: A Safety Detection Model for Substation Operations with Fused Contextual Information

Abstract: Detecting and regulating compliance at substation construction sites is critical to ensure the safety of workers. The complex backgrounds and diverse scenes of construction sites, as well as the variations in camera angles and distances, make the object detection models face low accuracy and missed detection problems. In addition, the high complexity of existing models creates an urgent need for effective parameter compression techniques to facilitate deployment at the edge server. To cope with these challenges, this study proposes a safety protection detection algorithm that fuses contextual information for substation operation sites, which enhances multi-scale feature learning through a two-path downsampling (TPD) module to effectively cope with changes in target scales. Meanwhile, the Global and Local Context Information extraction (GLCI) module is utilized to optimize the key information learning and reduce the background interference. Furthermore, the C3GhostNetV2 unit is utilized in discerning the interconnections of far-off spatial pixels, while enhancing the network's expressive power and reducing the number of parameters and computational costs. The outcomes of the experiments indicate that the present model improves upon the mAP50 metric by 4.5% compared to the baseline model, and the accuracy of the checks and the recall have seen respective increases of 4.8% and 10.1%, while there has been a reduction in both the count of parameters and the floating-point calculations by 16.5% and 12.6% respectively, which proves the validity and practicability of the method.

Author 1: Bo Chen

Author 2: Hongyu Zhang

Author 3: Runxi Yang

Author 4: Lei Zhao

Author 5: Yi Ding

Keywords: Object detection; context information; electricity construction operation; model complexity; lightweight

Paper 98: Preprocessing and Analysis Method of Unplanned Event Data for Flight Attendants Based on CNN-GRU

Abstract: The data of unplanned flight attendant events has characteristics such as diversity and complexity, which pose great challenges to data preprocessing and analysis. This study proposes a preprocessing and analysis method for unplanned flight attendant event data based on Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU). Firstly, an efficient word vector tool is used to preprocess the raw data, improving its quality and consistency. Then, convolutional neural networks are taken to extract local features of the data, combined with gated loop units to capture dynamic changes in time series, thus achieving effective analysis and prediction of unplanned events in air crew. The results showed that in the 6-class task, the research model exhibited the highest accuracy at 99.24%, the lowest accuracy at 94.24%, and an average accuracy of 98.53%. The highest, lowest, and average accuracies in the 10-class task were 96.63%, 90.17%, and 93.21%, respectively. The performance of the research model was superior to support vector machine, K-nearest neighbor algorithm, and some advanced algorithms. This study provides a more accurate analysis tool for unplanned event data of flight attendants, to assist in the efficiency of aviation emergency event handling and improve aviation safety.

Author 1: Dongyang Li

Keywords: Convolutional neural network; gate recurrent units; air crew; unplanned events; data preprocessing; data analysis

Paper 99: CNN-BiGRU-Focus: A Hybrid Deep Learning Classifier for Sentiment and Hate Speech Analysis of Ashura-Arabic Content for Policy Makers

Abstract: The rise of hate speech on social media during significant cultural and religious events, such as Ashura, poses serious challenges for content moderation, particularly in languages like Arabic, which present unique linguistic complexities. Most existing hate speech detection models, primarily developed for English text, fail to effectively handle the intricacies of Arabic, including its diverse dialects and rich morphology. This limitation underscores the need for specialized models tailored to the Arabic language. In response, the CNN-BiGRU-Focus model proposed, a novel hybrid deep learning (DL) approach that combines Convolutional Neural Networks (CNN) to capture local linguistic patterns and Bidirectional Gated Recurrent Units (BiGRU) to manage long-term dependencies in sequential text. An attention mechanism is incorporated to enhance the model's ability to focus on the most relevant sections of the input, improving both the accuracy and interpretability of its predictions. In this paper, this model applied to a dataset of social media posts related to Ashura, revealing that 32% of the content comprised hate speech, with Shia users expressing more sentiments than Sunni users. Through extensive experiments, the CNN-BiGRU-Focus model demonstrated superior performance, significantly outperforming baseline models. It achieved an accuracy of 99.89% and AUC of 99, marking a substantial improvement in Ashura-Arabic hate speech detection. The model effectively addresses the linguistic challenges of Arabic, including dialect variations and nuanced contexts, making it highly suitable for content moderation tasks. To the best of author’s knowledge, this study represents the first attempt to compile an Arabic-Ashura hate detection dataset from Twitter and apply CNN-BiGRU-Focus DL model to detect hate sentiment in Arabic social media posts. Dataset and source code can be downloaded from (https://github.com/imamu-asa).

Author 1: Sarah Omar Alhumoud

Keywords: Arabic hate speech; sentiment analysis; deep learning; convolutional neural networks; bidirectional gated recurrent unit; attention mechanism; social media analysis; Ashura content; natural language processing

Paper 100: Percussion Big Data Mining and Modeling Method Based on Deep Neural Network Model

Abstract: In order to improve the analysis effector percussion waveform, this paper studies the percussion big data mining and modeling method based on the deep neural network model. Aiming at the problem of the high sampling rate of Analog to Digital Converter (ADC) when the wideband frequency-hopping Linear Frequency Modulation (LFM) percussion waveform is sampled by Nyquist, this paper proposes a method of under sampling, and conducts a simple theoretical analysis. When the signal-to-noise ratio is 35dB, the frequency measurement error is close to 1MHz, which can meet the requirements of frequency measurement accuracy. When the signal-to-noise ratio is higher than 35dB, the frequency measurement error gradually decreases and eventually stabilizes, with a frequency measurement accuracy of around 30 kHz. Due to the low environmental interference in the sound wave recognition of percussion instruments and the close distance between the hardware equipment and the percussion instruments in this paper, the recognition results of the model in this paper have high accuracy Compared with existing methods, this article is more reliable in identifying percussion sound waves. From the data, it can be seen that the method proposed in this article has better performance in waveform recognition in impact big data mining models.

Author 1: Xi Song

Keywords: Deep neural network; percussion; big data; mining; modeling

Paper 101: Deep Image Keypoint Detection Using Cascaded Depth Separable Convolution Modules

Abstract: Depth images have become an important data source for human bone keypoint detection due to their three-dimensional information. To optimize the efficiency of keypoint detection in depth images, a depth image keypoint detection model that combines cascaded depth separable convolution modules is constructed. The model first performs data cleaning and preprocessing on the image, replacing traditional convolutional layers with depthwise separable convolutional modules. The Faster OpenPose network is introduced to replace the traditional convolutional network structure with the lighter MobileNetV1 for detecting keypoints in the image. When the dataset size was 4000, the Faster OpenPose model had an accuracy of 0.97 and a mean square error of 0.03. The recognition rates for four different images were 0.91, 0.87, 0.89, and 0.93, respectively. The processing times were 0.32, 0.31, 0.28, and 0.27, respectively. The method of depth image keypoint detection combined with cascaded depth separable convolution modules has good practicality and excellent detection performance for various images, providing new ideas for future keypoint detection technology research.

Author 1: Rui Deng

Keywords: Depth image; DWCA; key point detection; OpenPose; cascade depth

Paper 102: Development of a Service Robot for Hospital Environments in Rehabilitation Medicine with LiDAR-Based Simultaneous Localization and Mapping

Abstract: This paper presents the development and evaluation of a medical service robot equipped with 3D LiDAR and advanced localization capabilities tailored for use in hospital environments. The robot employs LiDAR-based Simultaneous Localization and Mapping (SLAM) to navigate autonomously and interact effectively within complex and dynamic healthcare settings. A comparative analysis with the established 3D-SLAM technology in Autoware version 1.14.0, under a Linux ROS framework, provided a benchmark for evaluating our system's performance. The adaptation of Normal Distribution Transform (NDT) Matching to indoor navigation allowed for precise real-time mapping and enhanced obstacle avoidance capabilities. Empirical validation was conducted through manual maneuvers in various environments, supplemented by ROS simulations to test the system’s response to simulated challenges. The findings demonstrate that the robot's integration of 3D LiDAR and NDT Matching significantly improves navigation accuracy and operational reliability in a healthcare context. This study not only highlights the robot's ability to perform essential tasks with high efficiency but also identifies potential areas for further improvement, particularly in sensor performance under diverse environmental conditions. The successful deployment of this technology in a hospital setting illustrates its potential to support medical staff and contribute to patient care, suggesting a promising direction for future research and development in healthcare robotics.

Author 1: Sayat Ibrayev

Author 2: Arman Ibrayeva

Author 3: Bekzat Amanov

Author 4: Serik Tolenov

Keywords: Medical service robots; 3D LiDAR technology; autonomous navigation; hospital environments; robot-assisted healthcare; healthcare robotics; operational reliability; patient care automation

Paper 103: Multi-Sensor Data Fusion Analysis for Tai Chi Action Recognition

Abstract: The continuous development of action recognition technology can capture the decomposition data of Tai Chi movements, provide precise assistance for learners to correct erroneous movements and enhance their interest in practicing Tai Chi. Inertial sensors and human skeletal models are used to collect motion data. Combined with visual sensors, the motion and trajectory of Tai Chi are processed to obtain the relevant coordinate system of the movement trajectory. Then, the inertial sensor and visual sensor are fused for data processing to standardize the human skeleton model, remove noise interference from the collected information, and improve the smoothness performance of movement trajectories, thereby segmenting and clustering Tai Chi movement trajectories. Finally, the support vector machine and dynamic time-warping algorithm are combined to identify and verify the trajectory of Tai Chi movements. According to the results, in the 25%, 50%, and 75% training sample proportions, the lowest recognition accuracy of the Qi Shi movements was 90.87%, 93.53%, and 98.08%, respectively. The optimal recognition accuracy and standard deviation of single nodes in binary classification were 98.48% and 0.47%, respectively. The best recognition accuracy and standard deviation for multi-joint points in binary classification were 99.77% and 0.16%, respectively. This proves the recognition advantages of binary classification and the superiority of data fusion analysis based on multiple sensors, providing a theoretical basis and technical reference for action recognition technology.

Author 1: Jingying Ouyang

Author 2: Jisheng Zhang

Author 3: Yuxin Zhao

Author 4: Changhuo Yang

Keywords: Inertial sensor; visual sensors; segmentation clustering; support vector machine; dynamic time warping algorithm

Paper 104: Skywatch: Advanced Machine Learning Techniques for Distinguishing UAVs from Birds in Airspace Security

Abstract: This study addresses the critical challenge of distinguishing Unmanned Aerial Vehicles (UAVs) from birds in real-time for airspace security in both military and civilian contexts. As UAVs become increasingly common, advanced systems must accurately identify them in dynamic environments to ensure operational safety. We evaluated several machine learning algorithms, including K-Nearest Neighbors (kNN), AdaBoost, CN2 Rule Induction, and Support Vector Machine (SVM), employing a comprehensive methodology that included data preprocessing steps such as image resizing, normalization, and augmentation to optimize training on the "Birds vs. Drone Dataset." The performance of each model was assessed using evaluation metrics such as accuracy, precision, recall, F1 score, and Area Under the Curve (AUC) to determine their effectiveness in distinguishing UAVs from birds. Results demonstrate that kNN, AdaBoost, and CN2 Rule Induction are particularly effective, achieving high accuracy while minimizing false positives and false negatives. These models excel in reducing operational risks and enhancing surveillance efficiency, making them suitable for real-time security applications. The integration of these algorithms into existing surveillance systems offers robust classification capabilities and real-time decision-making under challenging conditions. Additionally, the study highlights future directions for research in computational performance optimization, algorithm development, and ethical considerations related to privacy and surveillance. The findings contribute to both the technical domain of machine learning in security and broader societal impacts, such as civil aviation safety and environmental monitoring.

Author 1: Muhyeeddin Alqaraleh

Author 2: Mowafaq Salem Alzboon

Author 3: Mohammad Subhi Al-Batah

Keywords: Unmanned Aerial Vehicles (UAVs); machine learning; image recognition; real-time processing; security; computer vision; image processing

Paper 105: Design and Research of Artwork Interactive Exhibition System Based on Multi-Source Data Analysis and Augmented Reality Technology

Abstract: The current system has problems such as low efficiency of data processing, lack of smooth user experience and poor combination of display content and interactive technology, etc. There is a pressing need to optimize the integration of data analysis and augmented reality technology to improve the interactivity and visual appeal of exhibitions. This paper introduces and validates a combined prediction model based on multi-source data from the Internet. When using speeded-up robust features (SURF)-64 with a threshold of 500, the number of feature matches is 800, and the matching time is 162.85 ms. At a threshold of 1000, the number of matches drops to 510, and the time decreases to 96.54 ms. For SURF-128, the corresponding matches were 763 and 496, with times of 208.63 ms and 134.21 ms. This indicates that increasing the threshold not only reduces the number of matches but also shortens the matching time, likely due to fewer feature points simplifying the matching process.

Author 1: Xiao Chen

Author 2: Qibin Wang

Keywords: Multi-source data; feature analysis; augmented reality technology; artwork interactive exhibition system; prediction model

Paper 106: Optimization of Carbon Dioxide Dense Phase Injection Model Based on DBN Deep Learning Algorithm

Abstract: Carbon dioxide dense phase injection images have providing new research ideas for differential detection. Aiming at the drawbacks of large data volume, low matching efficiency, and longtime consumption of high-resolution carbon dioxide dense phase injection models, a registration algorithm for carbon dioxide dense phase injection models based on quadratic matching is proposed. This algorithm first uses down sampling to reduce image dimensions. A difference detection algorithm based on weakly supervised deep confidence network is proposed to neural networks, as well as the high manual labeling workload, low efficiency, and insufficient labeled data of high-resolution carbon dioxide dense phase injection models. This article first explores the throttling of CO2 venting in pipelines through the analysis of CO2 phase equilibrium characteristics. The experiment shows that there is after the valve, the greater the temperature drop. At the same time, water content will affect the throttling temperature drop is about 1.5 degrees; when the gas-liquid ratio is 2500, the throttling temperature drop is 7.4 degrees. CO2 in the reactor to over 8MPa, achieving supercritical pressure. CO2 with the constant temperature water bath is 5~100 degrees, with a temperature control accuracy of ± 0.1 degrees. The temperature of the water inside the water bath jacket of the kettle is adjusted through circulation. The maximum pressure of the kettle is 25MPa and the volume is 6L.

Author 1: Juan Zhou

Author 2: Dalong Wang

Author 3: Tieya Jing

Author 4: Zhiwen Liu

Author 5: Yihe Liang

Author 6: Yaowu Nie

Keywords: Supercritical CO2; DBN deep learning algorithm; throttling characteristics; security control; dense phase injection model

Paper 107: Intelligent Medical Multi-Department Information Attribute Encryption Access Control Method Under Cloud Computing

Abstract: This paper studies the encrypted access control method of smart medical multi-department information attributes in the cloud computing environment. Under the current wave of informatization, smart medical care has become an important development direction of medical services. However, the ensuing information security issues have become increasingly prominent. Especially in the context of cloud computing, information sharing and cooperative work among multiple departments make information security access control particularly important. This article first conducts an in-depth analysis of the characteristics of multi-departmental information in smart medical care. By collecting and sorting out a large amount of medical data, it is found that more than 85% of the data involves patient privacy and core information of medical business, which puts forward extremely high requirements for data confidentiality and integrity. Therefore, we propose an attribute-based encrypted access control method, which realizes differentiated access control for different users and different departments through a fine division of data attributes. In the implementation of the method, we design efficient encryption and decryption algorithms by using the distributed characteristics of cloud computing. Experimental results show that, compared with traditional access control methods, the proposed method improves the efficiency of data processing while ensuring information security, and the average access response time is reduced by about 20%. In addition, we also verified the effectiveness of this method in multi-department information security management of smart medical care through actual case analysis.

Author 1: Shubin Liao

Keywords: Cloud computing; smart medical care; multi-sectoral information; attribute encryption

Paper 108: Application of Data Exchange Model and New Media Technology in Computer Intelligent Auxiliary Platform

Abstract: To improve the use of computer-assisted learning, the author presents a method based on the use of new technologies. The system hardware model has a three-layer model system, including the user interface layer, the business option layer, and the data management layer. After teachers, students, and other users log in to the system, the user interface layer needs to enter personal information and advise users by including business-level options. System interaction is mainly influenced by two things: interactive learning and information sharing, interactive learning affects the online lessons of teachers and students; Data exchange is reflected in the data transmission of the data exchange model. According to the test results, after using the system, students with high self-efficacy increased from 11 to 21, and the percentage increased from 21.4 to 34.8, which can be understood as an increase in the number of good students. This interactive learning has been proven to increase students' self-efficacy and improve learning, and the use of this system has been a positive outcome.

Author 1: Na Li

Keywords: Computer intelligent assisted teaching system; information exchange; stress testing; new media technology

Paper 109: Blockchain-Enhanced Security and Efficiency for Thailand’s Health Information System

Abstract: This study seeks to enhance the security, efficiency, and usability of Thailand’s health information system through the integration of blockchain technology and a user-friendly web application. Blockchain’s inherent strengths in secure data stor-age and sharing make it particularly well-suited for addressing the critical challenges of healthcare data management. Currently, Thai citizens face significant barriers when seeking medical treatment across multiple hospitals, as they must manually request and transfer both their Electronic Medical Records (EMRs) and paper-based medical records. This fragmented system creates delays and inefficiencies, as each hospital operates its own isolated data silo. To overcome these challenges, this study proposes a solution utilising a private blockchain to securely store and manage patient medical histories and prescriptions. This approach ensures data integrity while implementing robust authorisation mechanisms to restrict access to sensitive information exclusively to verified individuals. The system’s security is further strengthened by blockchain’s encryption features and the use of smart contracts. A well-designed web application serves as the interface between the secure blockchain database and end-users, offering a seamless experience for both healthcare providers and patients. In addition, User Experience (UX) testing was conducted with healthcare providers to assess the system’s usability and functionality. The results highlight the system’s user-friendly interface, confirming its potential for widespread adoption. By fostering efficient, secure, and patient-centric health information exchange, this study has the potential to significantly enhance healthcare delivery and outcomes in Thailand.

Author 1: Thattapon Surasak

Keywords: Blockchain technology; healthcare information system; Electronic Medical Records (EMRs); user experience (UX) testing; web application; data security

Paper 110: Automatic Generation of Comparison Charts of Similar GitHub Repositories from Readme Files

Abstract: GitHub is a widely used platform for hosting open-source projects, with over 420 million repositories, promoting code sharing and reusability. However, with this tremendous number of repositories, finding a desirable repository based on user needs takes time and effort, especially as the number of candidate repositories increases. A user search can result in thousands of matching results, whereas GitHub shows only basic information about each repository. Therefore, users evaluate repositories’ applicability to their needs by inspecting the documentation of each repository. This paper discusses how comparison charts of similar repositories can be automatically generated to assist users in finding the desirable repository, reducing the time required to inspect their readme files. First, we implement an unsupervised, keyword-driven classifier based on the Lbl2TransformerVec algorithm to classify relevant content of GitHub readme files. The classifier was trained on a dataset of readme files collected from Java, JavaScript, C#, and C++ repositories. The classifier is evaluated against a different dataset of readme files obtained from Python repositories. Evaluation results indicate an F1 score of 0.75. Then, we incorporate rule-based adjustments to enhance classification results by 13%. Finally, the unique features, similarities, and limitations are automatically extracted from readme files to generate comparison charts using Large Language Models (LLMs).

Author 1: Emad Albassam

Keywords: Multi-class classification; keyword-driven classification; rule-based classification; unsupervised classification; GitHub repositories; comparison charts

Paper 111: An Ontology-Based Intelligent Interactive Knowledge Interface for Groundnut Crop Information

Abstract: This paper presents an ontology-based interactive interface designed to provide farmers in Gujarat with information related to groundnut crops. An ontology specific to the groundnut crop was developed and used to create a semantic question-answering (QA) interface. The proposed QA interface converts natural language question into SPARQL Query and provides answer using the backbone ontology. The overall performance of the system is at par with the existing semantic QA system. Overall accuracy of QA System is 80%.

Author 1: Purvi H. Bhensdadia

Author 2: C. K. Bhensdadia

Keywords: Agriculture ontology; ontology construction; question answer system; groundnut ontology

Paper 112: Leveraging Semi-Supervised Generative Adversarial Networks to Address Data Scarcity Using Decision Boundary Analysis

Abstract: Convolutional Neural Networks (CNNs) are widely regarded as one of the most effective solutions for image classification. However, developing high-performing systems with these models typically requires a substantial number of labeled images, which can be difficult to acquire. In image classification tasks, insufficient data often leads to overfitting, a critical issue for deep learning models like CNNs. In this study, we introduce a novel approach to addressing data scarcity by leveraging semi-supervised classification models based on Generative Adversarial Networks (SGAN). Our approach demonstrates significant improvements in both efficiency and performance, as shown by variations in the evolution of decision boundaries and overall accuracy. The analysis of decision boundaries is crucial, as it provides insights into the model’s ability to generalize and effectively classify new data points. Using the MNIST dataset, we show that our approach (SGAN) outperform CNN methods, even with fewer labeled images. Specifically, we observe that the distance between the images and the decision boundary in our approach is larger than in CNN-based methods, which contributes to greater model stability. Our approach achieves an accuracy of 84%, while the CNN model struggles to exceed 72%.

Author 1: Mohamed Ouriha

Author 2: Omar El Mansouri

Author 3: Younes Wadiai

Author 4: Boujamaa Nassiri

Author 5: Youssef El Mourabit

Author 6: Youssef El Habouz

Keywords: Decision boundary; convolutional neural network; Generative Adversarial Networks; MNIST; classification; semi-supervised classification

Paper 113: Optimizing Wearable Technology Selection for Injury Prevention in Ice and Snow Athletes Using Interval-Valued Bipolar Fuzzy Programming

Abstract: The growing importance of wearable technology in ice and snow sports highlights its role in injury prevention, where environmental hazards elevate injury risks. To address this, we propose a decision-making model using interval-valued bipolar fuzzy programming (IVBFP) for the optimal selection of wearable devices focused on athlete safety. The model employs multi-criteria decision-making (MCDM) methods to evaluate critical factors such as comfort, safety, durability, and real-time monitoring. Fuzzy logic enhances the precision and consistency of decision-making. The IVBFP model addresses vital challenges, including the diverse performance metrics of wearable devices and the uncertainty in expert evaluations. In comparison analyses, the model exhibited a 15% enhancement in judgment accuracy and a 12% decrease in uncertainty relative to conventional techniques. The results underscore the model’s proficiency in correctly forecasting devices that mitigate injury risks, providing improved athlete protection. The approach effectively incorporates expert viewpoints and subjective evaluations, diminishing harm risk in simulated and actual datasets. This research is significant both theoretically and practically. It offers a comprehensive framework to guarantee athlete safety in extreme conditions, connecting scholars and practitioners.

Author 1: Aichen Li

Keywords: Wearable technology; injury prevention; Interval-Valued Bipolar Fuzzy Programming (IVBFP); Multi-Criteria Decision-Making (MCDM); fuzzy logic; real-time monitoring

Paper 114: LRSA-Hybrid Encryption Method Using Linear Cipher and RSA Algorithm to Conceal the Text Messages

Abstract: Computer science and telecommunications technologies have been experiencing rapid advancements in recent years to protect sensitive data or information from potential harm, misuse, or destruction. By enhancing data security through various methodologies and algorithms, data can be better protected against attacks that may compromise its confidentiality, particularly in the case of text messages. Linear cipher is one of the earliest forms of cryptographic systems which operates by shifting letters that may not provide the highest level of security but adds a layer of complexity to the initial encryption process. Rivest-Shamir-Adleman algorithm represents a more advanced and rigorous approach to encryption that resistant to more sophisticated attacks. The Rivest-Shamir-Adleman algorithm utilizes the mathematical properties of large prime numbers to establish a secure communication channel. The combination of both algorithms or hybrid algorithms employed for data security, the security of text messages is significantly improved, ensuring the confidentiality of the text messages during its transmission. Hence, this research proposes two types of hybrid algorithms, namely Gradatim LRSA and Optimized LRSA, which ensure the confidentiality of the text message using encryption and decryption processes. The results also show that the Optimized LRSA performs with less computation compared to the Gradatim LRSA.

Author 1: Rundan Zheng

Author 2: Chai Wen Chuah

Author 3: Janaka Alawatugoda

Keywords: Confidentiality; data encryption; hybrid encryption; linear cipher; RSA algorithm; Gradatim LRSA; Optimized LRSA

Paper 115: Enterprise Architecture Framework Selection for Collaborative Freight Transportation Digitalization: A Hybrid FAHP-FTOPSIS Approach

Abstract: Collaborative freight transportation plays a crucial role for Logistic Service Providers (LSPs) seeking to enhance profitability and service quality, yet it faces challenges at strategic, operational, and technical levels. Digital transformation creates opportunities to overcome these hurdles by extending collaboration beyond physical logistics to encompass information management and digital transformation. Enterprise Architecture Frameworks (EAFs) offer promising solutions by providing a holistic view of various levels within such ecosystems and ensuring alignment between information systems and strategic objectives. However, selecting the right EAF is a complex and critical step. This study introduces an innovative approach for selecting an enterprise Architecture (EA) framework to support the development of a collaborative freight transportation platform. It emphasizes the importance of adopting a systematic EA methodology in the digitalization of the freight transportation sector. The decision-making process integrates established techniques such as the Analytic Hierarchy Process (AHP) and the Fuzzy Technique for Order of Preference by Similarity to Ideal Solution (F-TOPSIS). Applied to a case study involving a Moroccan logistics company, the approach demonstrates effectiveness in framework selection. The study’s findings underscore the method’s significance as a valuable tool for organizations embarking on digital transformation through EA, offering adaptability across diverse industries and contexts.

Author 1: Abdelghani Saoud

Author 2: Adil Bellabdaoui

Author 3: Mohamed Lachgar

Author 4: Mohamed Hanine

Author 5: Imran Ashraf

Keywords: Digital transformation; freight transportation; enterprise architecture; multi-criteria decision-making; analytic hierarchy process; fuzzy technique for order of preference by similarity to ideal solution

Paper 116: ATG-Net: Improved Feature Pyramid Network for Aerial Object Detection

Abstract: Object detection in aerial images is gradually gaining wide attention and application. However, given the prevalence of numerous small objects in the Unmanned Aerial Vehicle (UAV) aerial images, the extraction of superior fusion features is critical for the detection of small objects. However, feature fusion in many detectors does not fully consider the specific characteristics of the detection task. To obtain suitable features for the detection task, the paper proposes an improved Feature Pyramid Network (FPN) named ATG-Net, which aims to improve the feature fusion capability. Firstly, we propose an Adaptive Tri-Layer Weighting (ATW) module that adaptively assigns weights to each layer of the feature map according to its size and content complexity. Secondly, a Triple Feature Encoding (TFE) module is implemented, which can fuse feature maps from three different scales. Finally, the paper incorporates the Global Attention Mechanism (GAM) into the network, which includes improved channel attention mechanisms and spatial attention mechanisms. The experiments are conducted on the VisDrone2020 dataset, and the result shows that the network significantly outperforms the baseline detector and a variety of popular object detectors, which significantly improves the feature fusion capability of the network and the detection accuracy of small objects.

Author 1: Junbao Zheng

Author 2: ChangHui Yang

Author 3: Jiangsheng Gui

Keywords: Object detection; feature pyramid network; adaptive tri-layer weighting; triple feature encoding; global attention mechanism

Paper 117: Deep Learning Classification of Gait Disorders in Neurodegenerative Diseases Among Older Adults Using ResNet-50

Abstract: Gait disorders in older adults, particularly those associated with neurodegenerative diseases such as Parkinson’s Disease, Huntington’s Disease, and Amyotrophic Lateral Sclerosis , present significant diagnostic challenges. Since these NDDs primarily affect older adults, it is crucial to focus on this population to improve early detection and intervention. This study aimed to classify these gait disorders in individuals aged 50 and above using vertical ground reaction force (vGRF) data. A deep learning model was developed, employing Continuous Wavelet Transform (CWT) for feature extraction, with data augmentation techniques applied to enhance dataset variability and improve model performance. ResNet-50, a deep residual network, was utilized for classification. The model achieved a validation accuracy of 95.06% overall, with class-wise accuracies of 97.14% for ALS vs CO, 92.11% for HD vs CO, and 93.48% for PD vs CO. These findings underscore the potential of combining vGRF data with advanced deep-learning techniques, specifically ResNet-50, to classify gait disorders in older adults accurately, a demographic critically affected by these diseases.

Author 1: K. A. Rahman

Author 2: E. F. Shair

Author 3: A. R. Abdullah

Author 4: T. H. Lee

Author 5: N. H. Nazmi

Keywords: Gait disorders; neurodegenerative diseases; deep learning; vertical Ground Reaction Force (vGRF); ResNet-50

Paper 118: How Predictable are Fitness Landscapes with Machine Learning? A Traveling Salesman Ruggedness Study

Abstract: The notion of fitness landscape (FL) has shown promise in terms of optimization. In this paper we propose a machine learning (ML) prediction approach to quantify FL ruggedness by computing the entropy. The approach aims to build a model that could reveal information about the ruggedness of unseen instances. Its contribution is attractive in many cases like black-box optimization and in case we can rely on the information of small instances to discover the features of larger and time-consuming ones. The experiment consists in evaluating multiple ML models for the prediction of the ruggedness of the traveling salesman problem (TSP). The results show that ML can provide, for instances of a similar problem, acceptable predictions and that it can help to estimate ruggedness of large instances in that case. However, the inclusion of several features is necessary to have a more predictable landscape, especially when dealing with different TSP instances.

Author 1: Mohammed El Amrani

Author 2: Khaoula Bouanane

Author 3: Youssef Benadada

Keywords: Fitness landscape analysis; optimization algorithms; machine learning; landscape ruggedness; traveling salesman problem

Paper 119: Analyzing EEG Patterns in Functional Food Consumption: The Role of PCA in Decision-Making Processes

Abstract: The impact of obesity and diabetes are two central reasons for the high rate of developing cardiovascular diseases in this country, which is largely due to their ultra-processed, diet-rich foods. Supervised Learning for Decision Making: A Case Study of Functional Food Taste Perceptions In this experiment, we trained ordinary consumers to estimate the taste preferences of a unique group from which no ratings were available(11) and established that decision making can be performed through supervision. A deep learning neural network architecture according to the present disclosure is designed to model the decision-making behavior of consumers consuming functional products. The efficiency of the model can be increased upto 1.23% making use of proper values for the rest of the hyperparameters as explained in experiments carried out where we set the optimal configuration so that nurturing it gives the best results.

Author 1: Mauro Daniel Castillo P´erez

Author 2: Jes´us Jaime Moreno Escobar

Author 3: Ver´onica de Jes´us P´erez Franco

Author 4: Ana Lilia Coria Pa´ez

Author 5: Oswaldo Morales Matamoros

Keywords: EEG analysis; functional foods; decision-making; deep learning; and Principal Component Analysis (PCA)

Paper 120: Learning Local Reconstruction Errors for Face Forgery Detection

Abstract: Although several deepfake detection technologies have achieved great detection accuracy inside the data domain in recent years, there are still limitations in cross-domain generalization. This is due to the model’s ease of fitting the data sample distribution in the training data domain and its tendency to detect a specific forgery trace in order to reach a judgment rather than catching generalized forgery traces. In this paper, we propose to learn Local Reconstruction Errors for face forgery detection. The local anomaly traces of the fake face are often mapped using the original real face as a reference; however, the original real face of the fake face cannot be acquired in the real scenario. Therefore, this solution designs a local reconstruction autoencoder trained with real samples. By masking key areas of the face, the original real face can be reconstructed. Because the autoencoder only learns how to restore the essential parts of the real face using local patches of real samples, it cannot recover the forging traces or target face information in the fake face. Therefore, the reconstructed image forms a reconstructed difference with the original image. This solution aids the model in detecting local differences in fake faces by producing feature-level local difference attention mappings in the network’s middle layer. A series of experiments demonstrate that this solution has good detection and generalization performance.

Author 1: Haoyu Wu

Author 2: Lingyun Leng

Author 3: Peipeng Yu

Keywords: Face forgery; deepfake detection; local anomalies; generalized detection

Paper 121: Integrated Detection and Tracking Framework for 3D Multi-Object Tracking in Vehicle-Infrastructure Cooperation

Abstract: Vehicle-infrastructure cooperative perception has emerged as a promising approach to enhance 3D multi-object tracking by leveraging complementary data from vehicle and infrastructure sensors. However, existing methods face significant challenges, including difficulty in handling occlusions, suboptimal identity association, and inefficiencies in trajectory management, limiting their performance in real-world scenarios. In this paper, we propose a novel vehicle-infrastructure cooperative 3D multi-object tracking framework that addresses these challenges through three key innovations. First, an integrated detection-tracking framework jointly optimizes object detection and tracking, enhancing temporal consistency and reducing errors caused by separately handling the two tasks. Second, the XIOU identity association metric leverages 3D spatial and geometric relation-ships, ensuring robust object matching even under occlusions. Third, a four-stage cascade matching (FSCM) strategy adaptively manages trajectories by leveraging detection and prediction confidences, enabling accurate tracking in complex environments. Evaluated on the V2X-Seq dataset, our method achieves a MOTA of 57.23 and a MOTP of 74.64, significantly reducing identity switches while ensuring low bandwidth consumption and reliable tracking, highlighting its effectiveness and suitability for real-world deployment.

Author 1: Tao Hu

Author 2: Ping Wang

Author 3: Xinhong Wang

Keywords: Vehicle-infrastructure cooperative perception; 3D multi-object tracking; XIOU metric; four-stage cascade matching; integrated detection-tracking framework

Paper 122: A Robust Model for a Healthcare System with Chunk Based RAID Encryption in a Multitenant Blockchain Network

Abstract: Healthcare informatics has revolutionized data extraction from large datasets. However, using analytics while protecting sensitive healthcare data is a major challenge. A novel methodology for Privacy-Preserving Analytics in Healthcare Records addresses this essential issue in this study. The multi-tenant Blockchain framework uses chunk-based RAID encryption. For the healthcare business, chunk-based RAID encryption in a multi-tenant blockchain architecture creates a durable, safe, and efficient solution for processing confidential healthcare information. This solution improves data security, integrity, availability, performance, regulatory compliance, and scalability by combining RAID and blockchain technology. Contemporary healthcare systems need these qualities to work well. This approach was done in Python, and the libraries used the VSCode tool. To maintain data security, integrity, and accessibility, a strong healthcare system architecture with chunk-based RAID encryption in a multi-tenant blockchain network requires various advanced technologies.

Author 1: Bharath Babu S

Author 2: Jothi K R

Keywords: Multi-tenant; chunk-based; RAID; blockchain; healthcare records

Paper 123: Enhanced Adaptive Hybrid Convolutional Transformer Network for Malware Detection in IoT

Abstract: Many university networks use IoT devices, which increases vulnerability and malware threats. The complex, multi-dimensional structure of IoT network traffic and the imbalance between benign and dangerous data make traditional malware detection techniques ineffective. The Adaptive Hybrid Convolutional Transformer Network (AHCTN) is a novel model that uses CNNs for spatial feature extraction and Transformer networks for global temporal dependencies in IoT data. Unique preprocessing methods like Category Importance Scaling and Logarithmic Skew Compensation handle unbalanced data and severely skewed numerical characteristics. The Unified Feature Selector combines statistical and model-based feature selection methods and guarantees that only the most relevant characteristics are utilized for classification. DWS and LRW handle data imbalance. Our feature engineering approaches, such as Flow Efficiency and Packet Interarrival Consistency, improve prediction accuracy by capturing essential data correlations. The integration of advanced machine learning techniques ensures precise malware classification and enhances cybersecurity by addressing vulnerabilities in IoT-driven academic networks. The AHCTN model was carefully tested using the IoEd-Net dataset, which contains a variety of IoT devices and network activity. The AHCTN outperforms previous models with 98.9% accuracy. It also performs well in Log Loss (0.064), AUC (99.1%), Weighted Temporal Sensitivity (97.1%), and Anomaly Detection Score (96.8%), recognizing uncommon but essential abnormalities in academic network data. These findings demonstrate AHCTN’s robustness and scalability for academic IoT malware detection.

Author 1: Abdulaleem Ali Almazroi

Keywords: IoT security; malware detection; convolutional transformer network; cybersecurity; machine learning; network anomaly detection

Paper 124: Enhanced State Monitoring and Fault Diagnosis Method for Intelligent Manufacturing Systems via RXET in Digital Twin Technology

Abstract: To maintain efficiency and continuity in Industry 4.0, intelligent manufacturing systems use enhanced problem detection and condition monitoring. Existing models typically miss uncommon and essential errors, causing expensive downtimes and lost production. ResXEffNet-Transformer (RXET), a hybrid deep learning model, improves defect identification and pre-dictive maintenance by integrating ResNet, Xception, Efficient-Net, and Transformer-based attention processes. The algorithm was trained on a five-year Texas industrial dataset using IoT-enabled gear and digital twins. To manage data imbalances and temporal irregularities, a strong preprocessing pipeline included Dynamic Skew Correction, Temporal Outlier Normalization, and Harmonic Temporal Encoding. The Adaptive Statistical Evolutionary Selector (ASES) optimized feature selection using the Stochastic Feature Evaluator (SFE) and Evolutionary Divergence Minimizer (EDM) to increase prediction accuracy. The RXET model beat traditional methods with 98.9% accuracy and 99.2%AUC. Two new performance metrics, Temporal Fault Detection Index (TFDI) and Fault Detection Variability Coefficient (FDVC), assessed the model’s capacity to identify problems early and consistently across fault kinds. Simulation findings showed the RXET’s superiority in anticipating uncommon but essential errors. Pearson correlation (0.93) and ANOVA (F-statistic: 8.52) validated the model’s robustness. The sensitivity study showed the best performance with moderate learning rates and batch sizes. RXET provides a complete, real-time problem detection solution for intelligent industrial systems, improving predictive maintenance and addressing challenges in Industry 4.0, digital twin technology, IoT, and machine learning. The proposed RXET model enhances operational reliability in intelligent manufacturing and sets a foundation for future advancements in predictive analytics and large-scale industrial automation.

Author 1: Min Li

Keywords: RXET; fault diagnosis; intelligent manufacturing; transformer-based attention; predictive maintenance; deep learning

Paper 125: Recognizing Multi-Intent Commands of the Virtual Assistant with Low-Resource Languages

Abstract: Virtual Assistants (VAs) are widely used in many fields. Recently, VAs have been effectively applied in technical drawing tasks, such as in Photoshop and Microsoft Word. Understanding multi-intent commands in VAs poses a significant challenge, especially when the language in query is low-resource, like Vietnamese (no training dataset available for technical drawing domain), which features complex grammar and a limited domain of usage. In this work, we proposing a three-step process to develop a voice assistant capable of understanding multi-intent commands in VAs for low-resource languages, particularly in responding to the SCADA Framework (SF) for performing drawing tasks: (1) for the training dataset, we developed a semi-automatic method for building a labeled command corpus; applying this method to Vietnamese, we built a corpus that includes 3,240 labeled commands; (2) for the multi-intent command processing phase, we introduced a method for splitting multi-intent commands into single-intent commands to enable VAs to perform them more efficiently. By experimenting with the proposed method in Vietnamese, we developed a VA that supports drawing on SF with an accuracy of over 96%. With the results of this study, we can completely apply them to SCADA system products to support the automatic control of techinical drawing operations in them as VAs.

Author 1: Van-Vinh Nguyen

Author 2: Ha Nguyen-Tien

Author 3: Anh-Quan Nguyen-Duc

Author 4: Trung-Kien Vu

Author 5: Cong Pham-Chi

Author 6: Minh-Hieu Pham

Keywords: Vietnamese command corpus; chatbot; virtual assistants; multi-intent command; artificial intelligence; technical drawing; SCADA framework; build semi-automatic data; low-resource languages

Paper 126: AudioMag: A Hybrid Audio Protection Scheme in Multilevel DWT-DCT-SVD Transformations for Data Hiding

Abstract: Steganography is a technique used to hide data within an image or audio in order to maintain the secrecy of the message being communicated. There are several methods used in steganography to achieve this, but commonly, the data hiding is between the same stegoentity, such as an image with an image or audio with audio. One drawback of hiding data within the same entity is that once the security is compromised, one may be able to access the particular data. Therefore, this research proposes hiding audio within an image. The first step is to transform the audio into a hexadecimal value. Next, the hexadecimal value is hidden within the shortened uniform resource locators. The uniform resource locators are concatenated, shuffled, and converted into a quick response code. Finally, the quick response code is embedded into an image. The simulation results show the successful hiding of the audio message within the image, while maintaining the security and confidentiality of the hidden messages.

Author 1: Jingjin Yu

Author 2: Chai Wen Chuah

Author 3: Rundan Zheng

Author 4: Janaka Alawatugoda

Keywords: Steganography; image steganography; audio steganography; data hiding; stegoentity

Paper 127: Using Hybrid Compact Transformer for COVID-19 Detection from Chest X-Ray

Abstract: By the end of December 2019, the novel coronavirus 2019 (COVID-2019), became a world pandemic affecting the respiratory system. Scientists started investigating using Deep Learning and Convolutional Neural Networks (CNNs) to detect COVID-19 using Chest X-rays (CXRs). One of the main difficulties researchers reported in the detection of lung diseases is the fact that radiographic images can tell that the lungs are abnormal, but they might miss specifying the type of pneumonia exactly. Only the expert radiologist can tell the difference based on patches shapes and distribution on the affted lungs. Also CNN’s require big datasets to provide good results. When new pandemics spread, The limited benchmark datasets for COVID- 19 in CXR images, especially during the onset of the pandemic, is the main motivation of this research. In this research, we will introduce the use of Vision Transformers (ViTs). We consider an updated version of ViT called Compact Transformer (CT) which was proposed to reduce the expansive computations of the self-attention mechanism in ViT and to escape the big data paradigm. As a contribution of this study, We propose using a Hybrid Compact Transformer (HCT) in which a pretrained CNN is used in place of the convolutional layers in CT. Hence, with the hybrid model design, we aim to combine the localization power of CNNs, with the generalization power (attention mechanism or distanced-pixel relations) of ViTs. Based on experimental results using different performance metrics, the Hybrid Compact Transformer is shown to be superior over Compact Transformers and Transfer Learning models. Our proposed technique enjoys the benefits of both worlds; a faster training of the model due to TL with CNNs and reduced data requirements due to CT. Combining localized filters of CNNs and the attention mechanism of CT seems to provide a better discrimination between common pneumonia and Covid-19 pneumonia.

Author 1: Ghadeer Almoeili

Author 2: Abdenour Bounsiar

Keywords: Deep convolutional neural network; CXR chest X-Ray; COVID-19 pneumonia; vision transformers; compact convolutional transformer; hybrid compact transformer

Paper 128: AI-Blockchain Approach for MQTT Security: A Supply Chain Case Study

Abstract: The use of the MQTT protocol in critical sectors such as healthcare and industry has prompted research to propose solutions for strengthening its security and preventing it from attacks that are growing exponentially and becoming increasingly sophisticated and difficult to detect. This paper aims to improve the security of the MQTT architecture, ensuring it is resilient to current attacks and adaptable to potential future attacks while considering the constraints of the IoT environment. To achieve this, the proposed architecture is based on the interaction between the AI model, which continuously analyzes device behavior, and smart contracts, which automatically apply appropriate security measures once fraud is detected. A device reputation mechanism is used to prevent malicious devices from rejoining the network. The AI model proposed in this article was initially trained on a set of malicious behaviors using supervised learning. The results show that the detection accuracy achieved 95.97%. This accuracy is expected to improve over time through the integration of un-supervised learning into the architecture, enabling the discovery of new attack patterns and additional parameters for malicious behavior identification. For simulation testing, the architecture was applied to supply chain management as a case study of critical applications, and smart contracts were deployed in the Remix environment. The architecture demonstrated resilience and robustness across various attack scenarios.

Author 1: Raouya AKNIN

Author 2: Hind El Makhtoum

Author 3: Youssef Bentaleb

Keywords: IoT; MQTT; blockchain; smart contracts; AI hybrid model; device reputation

Paper 129: Optimized SMS Spam Detection Using SVM-DistilBERT and Voting Classifier: A Comparative Study on the Impact of Lemmatization

Abstract: The rapid growth of digital communication has led to a surge in spam messages, particularly through Short Message Service (SMS). These unsolicited messages pose risks such as phishing and malware, necessitating robust detection mechanisms. This study focuses on a comparative analysis of machine learning models for SMS spam detection, with a particular emphasis on a proposed SVM-DistilBERT model enhanced by a voting classifier. Using the UCI SMS Spam dataset, the models are evaluated based on recall, accuracy, precision, and Receiver Operating Characteristic Area Under the Curve (ROC AUC) scores to assess their effectiveness in correctly identifying spam messages. By leveraging Optuna for hyperparameter optimization, the proposed model achieves superior performance, with an accuracy of 99.6%, surpassing traditional methods like SVM with TF-IDF Bi-gram and AdaBoost, which achieved 98.03%. The study also examines the effects of lemmatization and synonym data augmentation, with lemmatization shown to improve spam detection by reducing feature space redundancy and enhancing semantic understanding. To ensure transparency in decision-making, Local Interpretable Model-Agnostic Explanations (LIME) is applied. The results demonstrate that the optimized SVM-DistilBERT with the voting classifier offers a robust and effective solution for SMS spam filtering.

Author 1: Sinar Nadhif Ilyasa

Author 2: Alaa Omar Khadidos

Keywords: SMS spam detection; Support Vector Machine (SVM); DistilBERT; hyperparameter optimization; LIME

Paper 130: A Machine Learning Approach to pH Monitoring: Mango Leaf Colorimetry in Aquaculture

Abstract: Maintaining optimal water quality is crucial for successful aquaculture. This necessitates careful management of various water quality parameters, including pH levels within their ideal range. There is growing interest in creating affordable optical pH sensors that provide accurate readings across a wide range of pH values. Development of sensors that are both accurate and cost-effective remains a challenge. To this end, this study demonstrates the use of machine learning with mango leaf extract as a colorimetric indicator to achieve accurate and cost-effective pH estimation for aquaculture practices. Mango leaf was utilized as the pH indicator, covering a range from 1 to 13. RGB color extraction and Exif data were used for image analysis to extract relevant features. The XGBoost algorithm, optimized through stepwise hyperparameter tuning with early stopping, was used to train three different models on this dataset to predict pH values. Three classification models, namely Y3, Y5, and Y13, were trained with 3, 5, and 13 output classes, respectively. The overall precision achieved by each model was 0.94, 0.85, and 0.72, respectively. This demonstrates the potential of this approach for developing a user-friendly yet cost-effective sensor for pH detection applicable in aquaculture practices. The proposed method could help aquaculture farmers an affordable and intelligent smartphone-based pH detection tool, enhancing water quality management while reducing the need for expensive instruments and eliminating the need for additional costly and time-consuming experimental work, thereby contributing to the sustainability of aquaculture practices.

Author 1: Hajar Rastegari

Author 2: Romi Fadilah Rahmat

Author 3: Farhad Nadi

Keywords: Aquaculture; machine learning; XGboost; water quality; sustainable aquaculture practices; water quality monitoring; mango leaf extract

Paper 131: Unveiling Hidden Variables in Adversarial Attack Transferability on Pre-Trained Models for COVID-19 Diagnosis

Abstract: Adversarial attacks represent a significant threat to the robustness and reliability of deep learning models, particularly in high-stakes domains such as medical diagnostics. Advanced Persistent Threat (APT) attacks, characterized by their stealth, complexity, and persistence, exploit adversarial examples to undermine the integrity of AI-driven healthcare systems, posing severe risks to their operational security. This study examines the transferability of adversarial attacks across pre-trained models deployed for COVID-19 diagnosis. Using two prominent convolutional neural networks (CNNs), ResNet50 and EfficientNet-B0, this study explores critical factors that influence the transferability of adversarial perturbations, a vulnerability that could be strategically exploited by APT attackers. By investigating the roles of model architecture, pre-training dataset characteristics, and adversarial attack mechanisms, this research provides valuable insights into the propagation of adversarial examples in medical imaging. Experimental results demonstrate that specific model architectures exhibit varying levels of susceptibility to adversarial transferability. ResNet50, with its deeper layers and residual connections, displayed enhanced robustness against adversarial perturbations, whereas EfficientNet-B0, due to its distinct feature extraction strategy, was more vulnerable to perturbations crafted using ResNet50’s gradients. These findings underscore the influence of architectural design on a model’s resilience to adversarial attacks. By advancing the understanding of adversarial robustness in medical AI applications, this study offers actionable guidelines for mitigating the risks associated with adversarial examples and emerging threats, such as APT attacks, in real-world healthcare scenarios.

Author 1: Dua’a Akhtom

Author 2: Manmeet Mahinderjit Singh

Author 3: Chew XinYing

Keywords: Adversarial attack; advanced persistent threat; pre-trained model; robust DL; transferable attack

Paper 132: Image Information Hiding Processing Based on Deep Neural Network Algorithm

Abstract: In order to more effectively hide and extract image information, a deep neural network-based algorithm and computer-aided image information hiding method is proposed. The hardware design of the system includes the selection of the main control chip, the design of the parallel processing structure, and the design of the Ethernet communication circuit; Software design includes an image information hiding module, an image information extraction module, and a carrier image processing module. The operation of the image information processing system based on the reversible information hiding algorithm is realized through the system hardware and software design. The experimental results show that the carrier image processing degree of the design system is much higher than that of the traditional system, and the maximum value can reach 91%, indicating that the carrier image processing performance of the design system is better. The scheme proposed in this paper can improve the security of secret information while ensuring the quality of dense image. Follow-up studies will continue to explore the combination of adversarial learning and various traditional embedding algorithms to further improve the concealment of graph-hiding algorithms.

Author 1: Zhe Zhang

Keywords: Image information hiding; neural networks; system design; image acquisition; information processing

Paper 133: Intelligent Digital Virtual Clothing Display System Based on LDA Mathematical Model

Abstract: In order to understand the intelligent digital virtual clothing display system based on mathematical models, the author proposes a research on an intelligent digital virtual clothing display system based on LDA mathematical models. The author first analyzes the realization of clothing matching function, and selects the cooperation between human skin color and clothing field as the influencing factors of clothing color matching and style matching based on expert knowledge and historical experience. Secondly, based on the different characteristics of different skin tones and the knowledge of clothing color matching, a set of clothing matching recommendation plans is presented to recommend suitable colors for users to refer to. Additionally, clothing style recommendations and choices are set, divided into upper and lower clothing, allowing users to choose more independently, the system itself also provides certain reference matching knowledge. Finally, the clothing matching rules were converted into computer image data, through analysis of the current market and existing research results, it was decided to implement a clothing matching display system based on VR technology, while providing recommended clothing matching solutions, a three-dimensional space was constructed to display clothing, allowing users to watch the effects of clothing matching according to their own choices, provide a new way for users in the clothing industry who have this demand.

Author 1: Zhao Wu

Author 2: Qingyuan He

Keywords: Mathematical model; virtual technology; clothing display

Paper 134: Enhancing Alzheimer's Detection: Leveraging ADNI Data and Large Language Models for High-Accuracy Diagnosis

Abstract: Alzheimer's disease (AD), the most common type of dementia, is expected to affect 152 million people by 2050, emphasizing the importance of early diagnosis. This study uses the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, combining cognitive tests, biomarkers, demographic details, and genetic data to build predictive models. Using large language models (LLMs), specifically ChatGPT 3.5, we achieved high classification accuracy, with ROC AUC values of 0.98 for cognitively normal (CN) individuals, 0.99 for dementia, and 0.98 for mild cognitive impairment (MCI). These findings show that LLMs can handle complex data quickly and accurately. By focusing on numerical and text-based data instead of just imaging, this method provides a cost-effective and accessible option for diagnosing AD. Adding genetic information improves the predictions, reflecting the important role of genetics in AD risk. This study highlights the potential of combining different types of data with advanced machine learning and LSTM to improve early AD diagnosis. Future research should explore more ways to combine data and test different machine learning models to further enhance diagnostic tools.

Author 1: Hassan Almalki

Author 2: Alaa O. Khadidos

Author 3: Nawaf Alhebaishi

Keywords: Alzheimer; dementia; LLMs; ChatGPT; LSTM

Paper 135: Visual Recognition and Localization of Industrial Robots Based on SLAM Algorithm

Abstract: The front-end feature matching module of traditional SLAM systems is characterized by sparse or dense feature points, it is difficult to generate accurate camera trajectory and scene reconstruction results, in response to this problem, the author studied a fast reconstruction algorithm for any path based on V-SLAM, by using improved feature matching algorithms to accurately match feature points, the accuracy of scene sparse reconstruction and camera trajectory recovery has been improved, the backend optimization thread adopts segmented optimization matching to reduce the computational burden of reconstruction, and the performance of the V-SLAM system was improved through parallel processing, the matching results and camera trajectory error comparison results showed that the improved V-SLAM algorithm can quickly recover camera trajectory and scene reconstruction, with the development of multi-sensor collaborative coupling and multi view fusion technology, the V-SLAM method proposed by the author can add virtual 3D objects to real scenes, and the V-SLAM system can extract feature points in the screen in real-time and detect planar objects in the scene, ensure that multiple virtual objects in the scene meet geometric consistency with the actual scene, in the experiment, two objects were added to the virtual scene, users can interactively scale objects and add them without being affected by camera movements, ensuring consistency between objects and the real scene.

Author 1: Wei Cui

Author 2: Yuefan Zhao

Author 3: Litao Sun

Keywords: SLAM algorithm; industrial robot; visual recognition; location

Paper 136: Optimizing Threat Intelligence Strategies for Cybersecurity Awareness Using MADM and Hybrid GraphNet-Bipolar Fuzzy Rough Sets

Abstract: Advanced threat detection systems are needed more than ever as cyber-attacks become more advanced. A novel cybersecurity model uses Bipolar Fuzzy Rough Sets, Graph Neural Networks, and dense network (BFRGD-Net) architectures to identify threats with unmatched accuracy and speed. The approach optimizes threat detection using Dynamic Range Realignment, anomaly-driven feature enhancement, and a hybrid feature selection strategy on a comprehensive Texas dataset of 66 months of real-world network activity. With 97.8% accuracy, 97.5% F1-score, and 98.3% AUC, BFRGD-Net sets new standards in the field. Threat Detection Sensitivity shows the model’s capacity to find uncommon, high-severity threats, while Balanced Risk Detection Efficiency provides fast, accurate threat detection. The model has strong correlations and the highest statistical metrics scores compared to other techniques. Extensive simulations demonstrate the model’s capacity to discern threat levels, attack kinds, and response techniques. BFRGD-Net revolutionizes cybersecurity by seamlessly merging cutting-edge machine learning with specific insights. Its advanced threat detection and classification engine reduces false negatives and enables proactive critical infrastructure protection in real-time. The model’s adaptability to various attack situations makes it vital for cybersecurity resilience in a digital environment.

Author 1: Qian Zhang

Keywords: Cybersecurity awareness; threat intelligence; MADM framework; BFRGD-Net; hybrid model; deep learning