IJACSA Volume 15 Issue 4 - thesai.org

IJACSA Volume 15 Issue 4

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: A Comparative Analysis of Traditional and Machine Learning Methods in Forecasting the Stock Markets of China and the US

Abstract: In the volatile and uncertain financial markets of the post-COVID-19 era, our study conducts a comparative analysis of traditional econometric models—specifically, the AutoRegressive Integrated Moving Average (ARIMA) and Holt's Linear Exponential Smoothing (Holt's LES)—against advanced machine learning techniques, including Support Vector Regression (SVR), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRU). Focused on the daily stock prices of the S&P 500 and SSE Index, the study utilizes a suite of metrics such as R-squared, RMSE, MAPE, and MAE to evaluate the forecasting accuracy of these methodologies. This approach allows us to explore how each model fares in capturing the complex dynamics of stock market movements in major economies like the U.S. and China amidst ongoing market fluctuations instigated by the pandemic. The findings reveal that while traditional models like ARIMA demonstrate strong predictive accuracy over short-term horizons, LSTM networks excel in capturing complex, non-linear patterns in the data, showcasing superior performance over longer forecast horizons. This nuanced comparison highlights the strengths and limitations of each model, with LSTM emerging as the most effective in navigating the unpredictable dynamics of post-pandemic financial markets. Our results offer crucial insights into optimizing forecasting methodologies for stock price predictions, aiding investors, policymakers, and scholars in making informed decisions amidst ongoing market challenges.

Author 1: Shangshang Jin

Keywords: Machine learning; Holt's LES; SVR; LSTM; GRU

Paper 2: Classification of Thoracic Abnormalities from Chest X-Ray Images with Deep Learning

Abstract: Most Chest X-Rays (CXRs) are used to spot the existence of chest diseases by radiologists worldwide. Examining multiple X-rays at the busiest medical facility may result in time and financial loss. Furthermore, in the detection of the disease, expert abilities and attention are needed. CXRs are usually used for the detection of heart and lung region anomalies. In this research, multi-level Deep Learning for CXRs ailment detection has been used to identify solutions to these issues. Spotting these anomalies with high precision automatically will significantly improve the processes of realistic diagnosis. However, the absence of efficient, public databases and benchmark analyses makes it hard to match the appropriate diagnosis techniques and define them. The publicly accessible VINBigData datasets have been used to address these difficulties and researched the output of established multi-level Deep Learning architectures on various abnormalities. A high accuracy in CXRs abnormality detection on this dataset has been achieved. The focus of this research is to develop a multi-level Deep Learning approach for Localization and Classification of thoracic abnormalities from chest radiograph. The proposed technique automatically localizes and categorizes fourteen types of thoracic abnormalities from chest radiographs. The used dataset consists of 18,000 scans that have been annotated by experienced radiologists. The YoloV5 model has been trained with fifteen thousand independently labeled images and evaluated on a test set of three thousand images. These annotations were collected via VinBigData's web-based platform, VinLab. Image preprocessing techniques are utilized for noise removal, image sequences normalization, and contrast enhancement. Finally, Deep Ensemble approaches are used for feature extraction and classification of thoracic abnormalities from chest radiograph.

Author 1: Usman Nawaz

Author 2: Muhammad Ummar Ashraf

Author 3: Muhammad Junaid Iqbal

Author 4: Muhammad Asaf

Author 5: Mariam Munsif Mir

Author 6: Usman Ahmed Raza

Author 7: Bilal Sharif

Keywords: Localization; classification; ensemble learning; YOLOV5; VINBigData; thoracic abnormalities; deep learning

Paper 3: Assisted Requirements Selection by Clustering using an Analytical Hierarchical Process

Abstract: This research investigates the fusion of the Analytic Hierarchy Process (AHP) with clustering techniques to enhance project outcomes. Two quantitative datasets comprising 20 and 100 software requirements are analyzed. A novel AHP dataset is developed to impartially evaluate clustering strategies. Five clustering algorithms (K-means, Hierarchical, PAM, GMM, BIRCH) are employed, providing diverse analytical tools. Cluster quality and coherence are assessed using evaluation criteria including the Dunn Index, Silhouette Index, and Calinski Harabaz Index. The MoSCoW technique organizes requirements into clusters, prioritizing critical requirements. This strategy combines strategic prioritization with quantitative analysis, facilitating objective evaluation of clustering results and resource allocation based on requirement priority. The study demonstrates how clustering can prioritize software requirements and integrate advanced data analysis into project management, showcasing the transformative potential of converging AHP with clustering in software engineering.

Author 1: Shehzadi Nazeeha Saleem

Author 2: Linda Mohaisen

Keywords: Requirements prioritization; next release plan; software product planning; decision support; MoSCoW; AHP; k-Means; GMM; BIRCH; PAM; hierarchical; clustering; clusters evaluation

Paper 4: Comparative Analysis of Telemedicine in Media Coverage Pre- and Post-COVID-19 using Unsupervised Latent Dirichlet Topic Modeling

Abstract: Telemedicine, driven by technology, has become a game-changer in healthcare, with the COVID-19 pandemic amplifying its significance by necessitating remote healthcare solutions. This study explores the evolution of telemedicine through news big data analysis. Our research encompassed a vast dataset from 51 media outlets (total 28,372 articles), including national and regional dailies, economic newspapers, broadcasters, and professional journals. Using LDA analysis, we delved into pre- and post-pandemic telemedicine trends comprehensively. A crucial revelation was the prominence of "medical law" in telemedicine discussions, underscoring the need for legal reforms. Keywords like "artificial intelligence" and "big data" underscored technology's pivotal role. Post-pandemic, keywords like "COVID-19," "online healthcare," and "telemedicine" surged, reflecting the pandemic's impact on remote healthcare reliance. These keywords' increased frequency highlights the pandemic's transformative influence. This study stresses addressing healthcare's legal constraints and maximizing technology's potential. To seamlessly integrate telemedicine, policy support and institutional backing are imperative. In summary, telemedicine's rise, propelled by COVID-19, signifies a healthcare paradigm shift. This study sheds light on its trajectory, emphasizing legal reforms, tech innovation, and pandemic-induced changes. The post-pandemic era must prioritize informed policy decisions for telemedicine's effective and accessible implementation.

Author 1: Haewon Byeon

Keywords: Telemedicine; COVID-19; medical law; healthcare transformation; LDA topic modeling

Paper 5: Distributed Optimization Scheduling Consistency Algorithm for Smart Grid and its Application in Cost Control of Power Grid

Abstract: There are problems such as low scalability and low convergence accuracy in the economic dispatch of smart grids. To address these situations, this study considers various constraints such as supply-demand balance constraints, climb constraints, and capacity constraints based on the unified consensus algorithm of multi-agent systems. By using Lagrange duality theory and internal penalty function method, the optimization of smart grid economic dispatch is transformed into an unconstrained optimization problem, and a distributed second-order consistency algorithm is proposed to solve the model problem. IEEE6 bus system testing showed that the generator cost of the distributed second-order consistency algorithm in the first, second, and third time periods was 2.2475 million yuan, 5.8236 million yuan, and 3.7932 million yuan, respectively. Compared to the first-order consistency algorithm, the generator cost during the corresponding time period has increased by 10.23%, 11.36%, and 13.36%. The actual total output has reached supply-demand balance in a short period of time with the changes in renewable energy, while maintaining supply-demand balance during the scheduling process. The actual total output during low, peak, and off peak periods was 99MW, 147MW, and 120MW, respectively. This study uses distributed second-order consistency algorithm to solve the economic dispatch model of smart grid to achieve higher convergence accuracy and speed. The study is limited by the assumption that the cost functions of each power generation unit are quadratic convex cost functions under ideal conditions. This economic dispatch model may not accurately reflect practical applications.

Author 1: Lihua Shang

Author 2: Meijiao Sun

Author 3: Cheng Pan

Author 4: Xiaoqiang San

Keywords: Distributed consistency algorithm; convex optimization; economic dispatch; smart grid

Paper 6: Evaluating the Accuracy of Cloud-based 3D Human Pose Estimation Tools: A Case Study of MOTiO by RADiCAL

Abstract: The use of 3D Human Pose Estimation (HPE) has become increasingly popular in the field of computer vision due to its various applications in human-computer interaction, animation, surveillance, virtual reality, video interpretation, and gesture recognition. However, traditional sensor-based motion capture systems are limited by their high cost and the need for multiple cameras and physical markers. To address these limitations, cloud-based HPE tools, such as DeepMotion and MOTiON by RADiCAL, have been developed. This study presents the first scientific evaluation of MOTiON by RADiCAL, a cloud-based 3D HPE tool based on deep learning and cloud computing. The evaluation was conducted using the CMU dataset, which was filtered and cleaned for this purpose. The results were compared to the ground truth using two metrics, the Mean per Joint Error (MPJPE) and the Percentage of Correct Keypoints (PCK). The results showed an accuracy of 98 mm MPJPE and 96% PCK for most scenarios and genders. This study suggests that cloud-based HPE tools such as MOTiON by RADiCAL can be a suitable alternative to traditional sensor-based motion capture systems for simple scenarios with slow movements and little occlusion.

Author 1: Hamza Khalloufi

Author 2: Mohamed Zaifri

Author 3: Abdessamad Benlahbib

Author 4: Fatima Zahra Kaghat

Author 5: Ahmed Azough

Keywords: 3D; human pose estimation; animation; evaluation; motion tracking

Paper 7: Optimizing Student Performance Prediction: A Data Mining Approach with MLPC Model and Metaheuristic Algorithm

Abstract: Given the information stored in educational databases, automated achievement of the learner’s prediction is essential. The field of educational data mining (EDM) is handling this task. EDM creates techniques for locating data gathered from educational settings. These techniques are applied to comprehend students and the environment in which they learn. Institutions of higher learning are frequently interested in finding how many students will pass or fail required courses. Prior research has shown that many researchers focus only on selecting the right algorithm for classification, ignoring issues that arise throughout the data mining stage, such as classification error, class imbalance, and high dimensionality data, among other issues. These kinds of issues decreased the model's accuracy. This study emphasizes the application of the Multilayer Perceptron Classification (MLPC) for supervised learning to predict student performance, with various popular classification methods being employed in this field. Furthermore, an ensemble technique is utilized to enhance the accuracy of the classifier. The goal of the collaborative approach is to address forecasting and categorization issues. This study demonstrates how crucial it is to do algorithm fine-tuning activities and data pretreatment to address the quality of data concerns. The exploratory dataset utilized in this study comes from the Pelican Optimization Algorithm (POA) and Crystal Structure Algorithm (CSA). In this research, a hybrid approach is embraced, integrating the mentioned optimizers to facilitate the development of MLPO and MLCS. Based on the findings, MLPO2 demonstrated superior efficiency compared to the other methods, achieving an impressive 95.78% success rate.

Author 1: Qing Hai

Author 2: Changshou Wang

Keywords: Educational data mining; multilayer perceptron classification; pelican optimization algorithm; crystal structure algorithm; student performance

Paper 8: DUF-Net: A Retinal Vessel Segmentation Method Integrating Global and Local Features with Skeleton Fitting Assistance

Abstract: Assisted evaluation through retinal vessel segmentation facilitates the early prevention and diagnosis of retinal lesions. To address the scarcity of medical samples, current research commonly employs image patching techniques to augment the training dataset. However, the vascular features in fundus images exhibit complex distribution, patch-based methods frequently encounter the challenge of isolated patches lacking contextual information, consequently resulting in issues such as vessel discontinuity and loss. Additionally, there are a higher number of samples with strong contrast vessels compared to those with weak contrast vessels in retinal images. Moreover, within individual patches, there are more pixels of strong contrast vessels compared to weak contrast vessels, leading to lower segmentation accuracy for small vessels. Hence, this study introduces a patch-based deep neural network method for retinal vessel segmentation to address the issues. Firstly, a novel architecture, termed Double U-Net with a Feature Fusion Module (DUF-Net), is proposed. This network structure effectively supplements missing contextual information and improves the problem of vessel discontinuity. Furthermore, an algorithm is introduced to classify vascular patches based on their contrast levels. Subsequently, conventional data augmentation methods were employed to achieve a balance in the number of samples with strong and weak contrast vessels. Additionally, method with skeleton fitting assistance is introduced to improve the segmentation of vessels with weak contrast. Finally, the proposed method is evaluated across four publicly available datasets: DRIVE, CHASE_DB1, STARE, and HRF. The results demonstrate that the proposed method effectively ensures the continuity of segmented blood vessels while maintaining accuracy.

Author 1: Xuelin Xu

Author 2: Ren Lin

Author 3: Jianwei Chen

Author 4: Huabin He

Keywords: Fundus image; vessel segmentation; skeleton fitting; data augmentation; patch classification

Paper 9: Novel Approaches for Access Level Modelling of Employees in an Organization Through Machine Learning

Abstract: In the contemporary business landscape, organizational trustworthiness is of utmost importance. Employee behavior, a pivotal aspect of trustworthiness, undergoes analysis and prediction through data science methodologies. Simultaneously, effective control over employee access within an organization is imperative for security and privacy assurance. This research proposes an innovative approach to model employee access levels using Geo-Social data and machine learning techniques like Linear Regression, K-Nearest Neighbours, Decision Tree, Random Forest, XGBoost, and Multi-Layered Perceptron. The data, sourced from social and geographical realms, encompasses details on employee geography, navigation preferences, spatial exploration, and choice set formations. Utilizing this information, a behavioral model is constructed to assess employee trustworthiness, categorizing them into access levels: low, moderate, high, and very high. The model's periodic review ensures adaptive access level adjustments based on evolving behavioral patterns. The proposed approach not only cultivates a more trustworthy organizational network but also furnishes a precise and reliable trustworthiness evaluation. This refinement contributes to heightened organizational coherence, increased employee commitment, and reduced turnover. Additionally, the approach ensures enhanced control over employee access, mitigating the risks of data breaches and information leaks by restricting the access of employees with lower trustworthiness.

Author 1: Priyanka C Hiremath

Author 2: Raju G T

Keywords: Access control; machine learning; employee behavior modeling; data analysis; organizational performance

Paper 10: Predicting Optimal Learning Approaches for Nursing Students in Morocco

Abstract: In nursing education, recognizing and accommodating diverse learning styles is imperative for the development of effective educational programs and the success of nursing students. This article addresses the crucial challenge of classifying the learning styles of nursing students in Morocco, where contextual studies are limited. To address this research gap, a contextual approach is proposed, aiming to develop a predictive model of the most appropriate learning approach (observational, experiential, reflective and active) for each nursing student in Morocco. This model incorporates a comprehensive set of variables such as age, gender, education, work experience, preferred learning strategies, engagement in social activities, attitudes toward failure, and self-assessment preferences. We used four multivariate machine learning algorithms, namely SVM, Tree, Neural Network, and Naive Bayes, to determine the most reliable and effective classifiers. The results show that neural network and decision tree classifiers are particularly powerful in predicting the most suitable learning approach for each nursing student. This research endeavors to enhance the success of nursing students and raise the overall quality of healthcare delivery in the country by tailoring educational programs to match individual learning styles.

Author 1: Samira Fadili

Author 2: Merouane Ertel

Author 3: Aziz Mengad

Author 4: Said Amali

Keywords: Learning styles; nursing students; predictive modeling; classification; personalized education

Paper 11: Automated Weeding Systems for Weed Detection and Removal in Garlic / Ginger Fields

Abstract: The global agriculture industry has faced various problems, such as rapid population growth and climate change. Among several countries, Japan has a declining agricultural workforce. To solve this problem, the Japanese government aims to realize “Smart agriculture” that applies information and communication technology, artificial intelligence, and robotics. Smart agriculture requires the development of robot technology to perform weeding and other labor-intensive agricultural tasks. Robotic weeding consists of an object detection method using machine learning to classify weeds and crops and an autonomous weeding system using robot hands and lasers. However, the approach used for these methods changes depending on the crop growth. The weeding system must consider the combination according to crop growth. This study addresses weed detection and autonomous weeding in crop-weed mixed ridges, such as garlic and ginger fields. We first develop a weed detection method using Mask R-CNN, which can detect individual weeds by instance segmentation from color images captured by an RGB-D camera. The proposed system can obtain weed coordinates in physical space based on the detected weed region and the depth image captured by the camera. Subsequently, we propose an approach to guide the weeding manipulator toward the detected weed coordinates. This paper integrates weed detection and autonomous weeding through these two proposed methods. We evaluate the performance of the Mask R-CNN trained on images taken in an actual field and demonstrate that the proposed autonomous weeding system works on a reproduced ridge with artificial weeds similar to garlic and weed leaves.

Author 1: Tsubasa Nakabayashi

Author 2: Kohei Yamagishi

Author 3: Tsuyoshi Suzuki

Keywords: Weed detection; weeding; mask R-CNN; agriculture robot

Paper 12: Enhancing Building Energy Efficiency: A Hybrid Meta-Heuristic Approach for Cooling Load Prediction

Abstract: The research tackles the complex problem of accurately predicting cooling loads in the context of energy efficiency and building management. It presents a novel approach that increases the precision of cooling load forecasts by utilizing machine learning (ML). The main objective is to incorporate a hybridization strategy into Radial Basis Function (RBF) models, a commonly used method for cooling load prediction, to improve their effectiveness. This new method significantly increases accuracy and reliability. The resulting hybrid models, which combine two powerful optimization techniques, outperform the state-of-the-art approaches and mark a major advancement in predictive modelling. The study performs in-depth analyses to compare standalone and hybrid model configurations, guaranteeing an unbiased and thorough performance evaluation. The deliberate choice of incorporating the Self-adaptive Bonobo Optimizer (SABO) and Differential Squirrel Search Algorithm (DSSA) underscores the significance of leveraging the distinctive strengths of each optimizer. The study delves into three variations of the RBF model: RBF, RBDS, and RRBSA. Among these, the RBF model, integrating the SABO optimizer (RBSA), distinguishes itself with an impressive R2 value of 0.995, denoting an exceptionally close alignment with the data. Furthermore, a low Root Mean Square Error (RMSE) value of 0.700 underscores the model's remarkable precision. The research showcases the effectiveness of fusing ML techniques in the RBSA model for precise cooling load predictions. This hybrid model furnishes more dependable insights for energy conservation and sustainable building operations, thereby contributing to a more environmentally conscious and sustainable future.

Author 1: Chenguang Wang

Author 2: Yanjie Zhou

Author 3: Libin Deng

Author 4: Ping Xiong

Author 5: Jiarui Zhang

Author 6: Jiamin Deng

Author 7: Zili Lei

Keywords: Building energy; cooling load; machine learning; radial basis function; self-adaptive bonobo optimizer; differential squirrel search algorithm

Paper 13: Securing IoT Environment by Deploying Federated Deep Learning Models

Abstract: The vast network of interconnected devices, known as the Internet of Things (IoT), produces significant volumes of data and is vulnerable to security threats. The proliferation of IoT protocols has resulted in numerous zero-day attacks, which traditional machine learning systems struggle to detect due to IoT networks' complexity and the sheer volume of these attacks. This situation highlights the urgent need for developing more advanced and effective attack detection methods to address the growing security challenges in IoT environments. In this research, we propose an attack detection mechanism based on deep learning for federated learning in IoT. Specifically, we aim to detect and prevent malicious attacks in the form of model poisoning and Byzantine attacks that can compromise the accuracy and integrity of the trained model. The objective is to compare the performance of a distributed attack detection system using a DL model against a centralized detection system that uses shallow machine learning models. The proposed approach uses a distributed attack detection system that consists of multiple nodes, each with its own DL model for detecting attacks. The DL model is trained using a large dataset of network traffic to learn high-level features that can distinguish between normal and malicious traffic. The distributed system allows for efficient and scalable detection of attacks in a federated learning network within the IoT. The experiments show that the distributed attack detection system using DL outperforms centralized detection systems that use shallow machine learning models. The proposed approach has the potential to improve the security of the IoT by detecting attacks more effectively than traditional machine learning systems. However, there are limitations to the approach, such as the need for a large dataset for training the DL model and the computational resources required for the distributed system.

Author 1: Saleh Alghamdi

Author 2: Aiiad Albeshri

Keywords: Internet of Things (IoT); security breaches; machine learning; Deep Learning (DL)

Paper 14: Review and Analysis of Financial Market Movements: Google Stock Case Study

Abstract: A financial marketplace where shares of companies with public listings are bought and sold is called the stock market. It serves as a gauge of a nation's economic health by taking into account the operations of individual businesses as well as the general business climate. The relationship between supply and demand affects stock prices. Though it might be dangerous, stock market investing has the potential to provide large rewards in the long run. Together with increased prediction accuracy, optimization techniques such as Biogeography-based optimization (BBO), Artificial bee colony algorithm (ABC) and Aquila Optimization (AO) Algorithm further enhance the Extreme gradient boosting (XGBoost) ability to adapt to changing market conditions. The results were 0.955, 0.966, 0.972, and 0.982 for XGBoost, BBO-XGBoost, ABC-XGBoost, and AO-XGBoost, in that order. The performance difference between AO-XGBoost and XGBoost shows how combining with the optimizer may enhance the model's performance. By comparing the output of many optimizers, the most accurate optimization has been determined to be the model's main optimizer.

Author 1: Yiming LU

Keywords: Stock future trend; financial market; investment; machine learning algorithms; Google stock

Paper 15: Prediction of Financial Markets Utilizing an Innovatively Optimized Hybrid Model: A Case Study of the Hang Seng Index

Abstract: Stock trading is a highly consequential and frequently discussed subject in the realm of financial markets. Due to the volatile and unpredictable nature of stock prices, investors are perpetually seeking methods to forecast future trends in order to minimize losses and maximize profits. Nevertheless, despite the ongoing investigation of various approaches to optimize the predictive efficacy of models, it is indisputable that a method for accurately forecasting forthcoming market trends does not yet exist. A multitude of algorithms are currently being employed to forecast stock prices due to significant developments that have occurred in recent years. An innovative algorithm for predicting stock prices are examined in this paper which is a Gated Recurrent Unit combined with the Aquila optimizer. A comprehensive data implementation utilizing the Hang Seng Index stock price was executed as a dataset of this research which was collected between the years of 2015 and the end of June 2023. In the study, several additional methods for predicting stock market movements are also detailed. A comprehensive comparative analysis of the stock price prediction performances of the aforementioned algorithms has also been carried out to offer a more in-depth analysis and then the results are displayed in an understandable tabular and graphical manner. The proposed model obtained the values of 0.9934, 0.71, 143.62, and 36530.58, for R^2, MAPE, MAE, and MSE, respectively. These results proved the efficiency and accuracy of the suggested method and it was determined that the proposed model algorithm produces results with a high degree of accuracy and performs the best when it comes to forecasting a time series or stock price.

Author 1: Xiaopeng YANG

Keywords: Financial markets; stock future trend; Hang Seng Index; Gated Recurrent Units; Aquila Optimizer

Paper 16: Research on Diagnosis Method of Common Knee Diseases Based on Subjective Symptoms and Random Forest Algorithm

Abstract: Knee diseases are common diseases in the elderly, and timely and effective diagnosis of knee diseases is essential for disease treatment and rehabilitation training. In this study, we construct a diagnostic model of common knee diseases based on subjective symptoms and random forest algorithm to realize patients' self-initial diagnosis. In this paper, we first constructed a questionnaire of subjective symptoms of knee, and set up a questionnaire system to guide users to fill out the questionnaire correctly. Then clinical data collection is carried out to obtain clinical questionnaire data. Finally, the diagnostic analysis of three common diseases of knee joint is carried out by random forest machine learning method. Through leave-one-out cross validation, the accuracy of meniscus injury, anterior cruciate ligament injury and knee osteoarthritis diseases are 0.79, 0.84, 0.81 respectively; the sensitivity is 0.79, 0.84, 0.88 respectively; and the specificity is 0.80, 0.84, 0.79 respectively. The results show that the method can achieve a good effect of self-diagnosis, and can provide a knee joint disease screening a convenient and effective approach.

Author 1: Guangjun Wang

Author 2: Mengxia Hu

Author 3: Linlin Lv

Author 4: Hanyuan Zhang

Author 5: Yining Sun

Author 6: Benyue Su

Author 7: Zuchang Ma

Keywords: Knee diseases; subjective symptoms; random forest algorithm; self-diagnosis

Paper 17: Data Dynamic Prediction Algorithm in the Process of Entity Information Search for the Internet of Things

Abstract: To address the issue of insufficient real-time capability in existing Internet search engines within the Internet of Things environment, this research investigates the architecture of Internet of Things search systems. It proposes a data dynamic prediction algorithm tailored for the process of entity information search in the Internet of Things. The study is based on the design of a data compression algorithm for the Internet of Things entity information search process using the Rotating Gate Compression Algorithm. The algorithm employs the Least Squares Support Vector Machine to dynamically predict changes in entity node states in the Internet of Things, aiming to reduce sensor node resource consumption and achieve real-time search. Finally, the research introduces an Internet of Things entity information search system based on the data dynamic algorithm. Performance test results indicate that the segmented compression algorithm designed in the study can enhance compression accuracy and compression rate. As compression accuracy increases, errors also correspondingly increase. The prediction algorithm designed in the study shows a decrease in node energy consumption as reporting cycles increase, reaching 0.2 at 5 cycles. At the 5-cycle point, the prediction errors on two research datasets are 0.5 and 7.8, respectively. The optimized data dynamic prediction algorithm in the study effectively reduces node data transmission, lowers node energy consumption, and accurately predicts node state changes to meet user search demands.

Author 1: Tianqing Liu

Keywords: Internet of Things; sensors; swinging door trending (SDT); support vector machine (SVM); data dynamic prediction

Paper 18: Deep Learning-Powered Lung Cancer Diagnosis: Harnessing IoT Medical Data and CT Images

Abstract: Currently, lung cancer poses a significant global threat, ranking among the most perilous and lethal ailments. Accurate early detection and effective treatments play pivotal roles in mitigating its mortality rates. Utilizing deep learning techniques, CT scans offer a highly advantageous imaging modality for diagnosing lung cancer. In this study, we introduce an innovative approach employing a hybrid Deep Convolutional Neural Network (DCNN), trained on both CT scan images and medical data retrieved from IoT wearable sensors. Our method encompasses a CNN comprising 22 layers, amalgamating latent features extracted from CT scan images and IoT sensor data to enhance the detection accuracy of our model. Training our model on a balanced dataset, we evaluate its performance based on metrics including accuracy, Area under the Curve (AUC) score, loss, and recall. Upon assessment, our method surpasses comparable approaches, exhibiting promising prospects for lung cancer diagnosis compared to alternative models.

Author 1: Xiao Zhang

Author 2: Xiaobo Wang

Author 3: Tao Huang

Author 4: Jinping Sheng

Keywords: Diagnosis; CT images; deep learning; convolutional neural network; lung cancer

Paper 19: Transmission Line Monitoring Technology Based on Compressed Sensing Wireless Sensor Network

Abstract: Given wireless sensor networks' significant data transmission requirements, conventional direct transmission often leads to bandwidth constraints and excessive network energy consumption. This paper proposes a transmission line monitoring technology based on compressed sensing wireless sensor networks to achieve real-time monitoring of ice-covered power lines. Grounded in compressed sensing theory, this method utilizes dual orthogonal wavelet transform sparse matrices for sparse representation of sensor data. Considering the practical requirements of power line monitoring, a data transmission model is established to implement compressed sampling transmission. The regularization orthogonal matching pursuit algorithm is employed for high-precision reconstruction of compressed data. The software and hardware components of the power line monitoring system are designed, and experiments are conducted under real-world conditions. The results demonstrate that: 1) the system operates stably with an ideal data compression effect, achieving a compression ratio of 93.191%. The absolute reconstruction errors for temperature, humidity, and wind speed sensor data are 0.064°C, 0.052%, and 0.128 m/s, respectively, indicating high reconstruction accuracy and effectively avoiding transmission impacts caused by bandwidth issues. 2) In a 36-hour energy consumption loss test, compared to direct transmission, the compressed transmission mode exhibits a lower rate of battery voltage decay, with a decrease of approximately 11.18%, effectively extending the network's lifespan.

Author 1: Shuling YIN

Author 2: Renping YU

Author 3: Longzhi WANG

Keywords: Compressed sensing; transmission line; wireless sensor network; orthogonal wavelet transform; data reconstruction

Paper 20: Implementation of Cosine Similarity Algorithm on Omnibus Law Drafting

Abstract: Drafting of Omnibus Laws presents a complex challenge in legal governance, often involving the integration and consolidation of disparate legal provisions into a unified framework. In this context, the application of advanced computational techniques becomes crucial for streamlining the drafting process and ensuring coherence across the law's various components. Cosine similarity, a widely used measure in natural language processing and document analysis, offers a quantitative means to assess the similarity between different sections or articles within the Omnibus Law draft. By representing legal texts as high-dimensional vectors in a vector space model, cosine similarity enables the comparison of textual similarity based on the cosine of the angle between these vectors. Implementing cosine similarity in the context of omnibus law using FastAPI and Laravel can be a valuable tool for analyzing similarity between legal documents, especially in the context of omnibus law. Legal practitioners and researchers can use the cosine similarity measure to compare the textual content of different legal documents and identify similarities. This can aid in tasks such as legal document retrieval, clustering similar provisions, and detecting potential inconsistencies. The combination of FastAPI and Laravel provides a potent and efficient way to develop and deploy this functionality, contributing to the advancement of legal informatics and analysis. The dataset used is Undang-Undang (UU) which used Bahasa from 1945 to 2022, comprising a total of 1705 UU. The implemented cosine similarity yielded a recall rate of 90.10% on the law.

Author 1: Aristoteles

Author 2: Muhammad Umaruddin Syam

Author 3: Tristiyanto

Author 4: Bambang Hermanto

Keywords: Cosine similarity; FastAPI; Laravel; Omnibus Law

Paper 21: Enhancing Particle Swarm Optimization Performance Through CUDA and Tree Reduction Algorithm

Abstract: In this paper, we present an enhancement for Particle Swarm Optimization performance by utilizing CUDA and a Tree Reduction Algorithm. PSO is a widely used metaheuristic algorithm that has been adapted into a CUDA version known as CPSO. The tree reduction algorithm is employed to efficiently compute the global best position. To evaluate our approach, we compared the speedup achieved by our CUDA version against the standard version of PSO, observing a maximum speedup of 37x. Additionally, we identified a linear relationship between the size of swarm particles and execution time; as the number of particles increases, so does computational load – highlighting the efficiency of parallel implementations in reducing execution time. Our proposed parallel PSOs have demonstrated significant reductions in execution time along with improvements in convergence speed and local optimization performance - particularly beneficial for solving large-scale problems with high computational loads.

Author 1: Hussein Younis

Author 2: Mujahed Eleyat

Keywords: Particle swarm optimization; tree reduction algorithm; parallel implementations; CUDA; GPU

Paper 22: Underwater Video Image Restoration and Visual Communication Optimization Based on Improved Non Local Prior Algorithm

Abstract: Underwater image processing should balance image clarity restoration and comprehensive display of underwater scenes, requiring image fusion and stitching techniques. The pixel level fusion method is based on pixels, and by fusing different image data, it eliminates stitching gaps and sudden changes in lighting intensity, preserves detailed information, and thus improves the accuracy of stitching images. In the process of restoring underwater video images without local priors, there is still room for optimization in steps such as removing atmospheric light values, estimating transmittance, and calculating dehazing images through regularization. Based on the characteristics of Jerlov water types, water quality is classified according to the properties of suspended solids, and each channel is adjusted to the compensation space to improve the restoration algorithm. Background light estimation is used to determine the degree of image degradation, select the optimal attenuation coefficient ratio, and restore the image. The experimental results show that it is crucial to choose a ratio of attenuation coefficients that is close to the actual water quality environment being photographed. Both this model and traditional algorithms have an accuracy rate of over 99.0%, with the accuracy of this model sometimes reaching 99.9%. Pixel level fusion and background light estimation technology optimize underwater images, improve stitching accuracy and clarity, enhance target detection and recognition, and have important value for marine exploration rigs.

Author 1: Tian Xia

Keywords: Improving non local prior algorithms; underwater video images; visual communication effect; optical characteristic processing; image quality

Paper 23: Integrated Ensemble Model for Diabetes Mellitus Detection

Abstract: Diabetes Mellitus, commonly referred to as (DM), is a chronic illness that affects populations worldwide, leading to more complications such as renal failure, visual impairment, and cardiovascular disease, thus significantly compromising the individual's well-being of life. Detecting DM at an early stage is both challenging and a critical procedure for healthcare professionals, given that delayed diagnosis can result to difficulties in managing the progression of the disease. This study seeks to introduce an innovative stacking ensemble model for early DM detection, utilizing an ensemble of machine learning and deep learning models. Our proposed stacking model integrates multiple prediction learners, including Random Forest (RF), Convolutional Neural Network (CNN) with Long Short-Term Memory networks (CNN-LSTM), and Sequential Dense Layers (SDLs) as base learner models, with the Extreme Gradient Boosting model (XGBoost) serving as the Meta-Learner model. Findings demonstrate that our proposed model achieves a 99% accuracy on the Pima dataset and 97% accuracy on the DPD dataset in detecting diabetes mellitus disease. In conclusion, our model holds promise for developing a diagnostic tool for DM disease, and it is recommended to conduct further testing on the types of diabetes mellitus to enhance and evaluate its performance comprehensively.

Author 1: Abdulaziz A Alzubaidi

Author 2: Sami M Halawani

Author 3: Mutasem Jarrah

Keywords: Diabetes mellitus; machine learning; deep learning; stacking; ensemble learning; RF; CNN-LSTM; SDLs; XGBoost

Paper 24: Leveraging Machine Learning Methods for Crime Analysis in Textual Data

Abstract: The proposed research paper explores the application of machine learning techniques in crime analysis problem, specifically focusing on the classification of crime-related textual data. Through a comparative analysis of various machine learning models, including traditional approaches and deep learning architectures, the study evaluates their effectiveness in accurately detecting and categorizing crime-related text data. The performance of the models is assessed using rigorous evaluation metrics, such as the area under the receiver operating characteristic curve (AUC-ROC), to provide insights into their discriminative power and reliability. The findings reveal that machine learning frameworks, particularly the deep learning model, consistently outperform conventional machine learning approaches, highlighting the potential of advanced neural network architectures in crime analysis tasks. The implications of these findings for law enforcement agencies and researchers are discussed, emphasizing the importance of leveraging advanced machine learning techniques to enhance crime prevention and intervention efforts. Furthermore, avenues for future research are identified, including the integration of multiple data sources and the exploration of interpretability and explainability of machine learning models in crime analysis problem. Overall, this research contributes to advancing the field of crime analysis problem and underscores the importance of leveraging innovative computational approaches to address complex societal challenges.

Author 1: Shynar Mussiraliyeva

Author 2: Gulshat Baispay

Keywords: Machine learning; artificial intelligence; crime analysis; text processing; natural language processing; text analysis; data-driven decision making

Paper 25: Superframe Segmentation for Content-based Video Summarization

Abstract: Video summarization is a complex computer vision task that involves the compression of lengthy videos into shorter yet informative summaries that retain the crucial content of the original footage. This paper presents a content-based video summarization approach that utilizes superframe segmentation to identify and extract keyframes representing the most significant information in a video. Unlike other methods that rely solely on visual cues, our approach segments the video into meaningful and coherent visual content units while also preserving the original video's temporal coherence. This method helps keep the context and continuity of the video in the summary. It involves dividing the video into superframes, each of which is a cluster of adjacent frames with similar motion and visual characteristics. The superframes are then ranked based on their salient scores, which are calculated using visual and motion features. The proposed method selects the top-ranked super frames for the video summary. It has been evaluated on the SUMMe and TVSum datasets and achieved state-of-the-art results for F1-score and accuracy. Based on the experimental outcomes, it is evident that the suggested superframe segmentation method is effective for video summarization, which could be largely assistive for monitoring and controlling the student activities, particularly during their online exams.

Author 1: Priyanka Ganesan

Author 2: Senthil Kumar Jagatheesaperumal

Author 3: Abirami R

Author 4: Lekhasri K

Author 5: Silvia Gaftandzhieva

Author 6: Rositsa Doneva

Keywords: Video summarization; deep learning; super frame segmentation; keyframes; keyshot identification

Paper 26: Training Model of High-Rise Building Project Management Talent under Multi-Objective Evolutionary Algorithm

Abstract: In order to meet the development needs of the construction engineering industry and further optimize and improve the talent training mode, this paper studies the talent training model of high-rise construction project management under the multi-objective evolutionary algorithm. The cognitive ability model of management talent is constructed, and the learning ability of management talent is analyzed. With the optimization objectives of minimizing the construction period, minimizing the project cost, and maximizing the benefit of skill growth in high-rise building projects, and taking the conditions of average proficiency and average duration of construction as constraints, the mixed immune genetic algorithm with the introduction of the double-island model is adopted to carry out multi-objective evolution of management talent training, so as to obtain the best training scheme for management talent in high-rise building projects. The experimental results show that after the optimization of this model, the skill proficiency of project management personnel can be effectively improved, construction time can be effectively reduced, construction efficiency can be improved, and construction costs can be improved.

Author 1: Pan QI

Keywords: Multi-objective evolution; high-rise building; engineering project; management personnel training; skill proficiency; project cost

Paper 27: Development of a New Chaotic Function-based Algorithm for Encrypting Digital Images

Abstract: This paper discusses the development of a new chaotic function (proposed chaotic map) as a keystream generator to be used to encrypt and decrypt the image. The proposed chaotic function is obtained through the composition process of two chaotic functions MS map and Tent map, with the aim of increasing data resistances to attacks. The randomness properties of the keystream generated by this function have been tested using Bifurcation diagrams, Lyapunov exponent, and NIST randomness analysis. All the analysis results indicate that the keystream passed the randomness tests and safe to be used for image encryption. The performance of the proposed chaotic function was measured by way of analysis of its initial value sensitivity, key space, and correlation coefficient of the encrypted image. This function can further increase the resilience against brute force attacks, minimizing the possibility of brute attacks, and has key combinations or key space of 1.05 × 10959 that is much greater than the key space generated by MS Map + Tent Map of 5.832 × 10958. Finaly, quantitative measurements of encrypted image quality show an MSE value of 0 and a PSNR value of ∞. These values mean that the encrypted image data is the same as its original and both are also visually identical.

Author 1: Dhian Sweetania

Author 2: Suryadi MT

Author 3: Sarifuddin Madenda

Keywords: Chaotic function; decryption; encryption; function composition; key space; MS tent map

Paper 28: Transfer Learning-based CNN Model for the Classification of Breast Cancer from Histopathological Images

Abstract: Breast cancer can have significant emotional and physical repercussions for women and their families. The timely identification of potential breast cancer risks is crucial for prompt medical intervention and support. In this research, we introduce innovative methods for breast cancer detection, employing a Convolutional Neural Network (CNN) architecture and Transfer Learning (TL) technique. Our foundation is the ICAIR dataset, encompassing a diverse array of histopathological images. To harness the capabilities of deep learning and expand the model's knowledge base, we propose a TL model. The CNN component adeptly extracts spatial features from histopathological images, while the TL component incorporates pretrained weights into the model. To tackle challenges arising from limited labeled data and prevent overfitting, we employ ResNet152v2. Utilizing a pre-trained CNN model on extensive image datasets initializes our CNN component, enabling the network to learn pertinent features from histopathological images. The proposed model achieves commendable accuracy (96.47%), precision (96.24%), F1-score (97.18%), and recall (96.63%) in identifying potential breast cancer cases. This approach holds the potential to assist medical professionals in early breast cancer risk assessment and intervention, ultimately enhancing the quality of care for women's health.

Author 1: Sumitha A

Author 2: Rimal Isaac R S

Keywords: Breast cancer; transfer learning; ResNet152v2; medical image analysis; ICAIR 2018 dataset

Paper 29: Autoencoder and CNN for Content-based Retrieval of Multimodal Medical Images

Abstract: Content-Based Medical Image Retrieval (CBMIR) is a widely adopted approach for retrieving related images by the comparison inherent features present in the input image to those stored in the database. However, the domain of CBMIR specific to multiclass medical images faces formidable challenges, primarily stemming from a lack of comprehensive research in this area. Existing methodologies in this field have demonstrated suboptimal performance and propagated misinformation, particularly during the crucial feature extraction process. In response, this investigation seeks to leverage deep learning, a subset of artificial intelligence for the extraction of features and elevate overall performance outcomes. The research focuses on multiclass medical images employing the ImageNet dataset, aiming to rectify the deficiencies observed in previous studies. The utilization of the CNN-based Autoencoder method manifests as a strategic choice to enhance the accuracy of feature extraction, thereby fostering improved retrieval results. In the ImageNet dataset, the results obtained from the proposed CBMIR model demonstrate notable average values for accuracy (95.87%), precision (96.03%) and recall (95.54%). This underscores the efficacy of the CNN-based autoencoder model in achieving good accuracy and underscores its potential as a transformative tool in advancing medical image retrieval.

Author 1: Suresh Kumar J S

Author 2: Maria Celestin Vigila S

Keywords: Medical image retrieval; multiclass medical images; artificial intelligence; deep learning; convolutional neural network; autoencoder

Paper 30: Optimizing Bug Bounty Programs for Efficient Malware-Related Vulnerability Discovery

Abstract: Conventional security measures struggle to keep pace with the rapidly evolving threat of malware, which demands novel approaches for vulnerability discovery. Although Bug Bounty Programs (BBPs) are promising, they often underperform in attracting researchers, particularly in uncovering malware-related vulnerabilities. This study optimizes BBP structures to maximize engagement and target malware vulnerability discovery, ultimately strengthening cyber defense. Employing a mixed-methods approach, we compared public and private BBPs and analyzed the key factors influencing researcher participation and the types of vulnerabilities discovered. Our findings reveal a blueprint for effective malware-focused BBPs that enable targeted detection, faster patching, and broader software coverage. This empowers researchers and fosters collaboration within the cybersecurity community, significantly reducing the attack surface for malicious actors. However, challenges related to resource sustainability and legal complexity persist. By optimizing BBPs, we unlocked a powerful tool to fight cybercrime.

Author 1: Semi Yulianto

Author 2: Benfano Soewito

Author 3: Ford Lumban Gaol

Author 4: Aditya Kurniawan

Keywords: Bug bounty; malware; vulnerability discovery; cyber defense

Paper 31: ConvADD: Exploring a Novel CNN Architecture for Alzheimer's Disease Detection

Abstract: Alzheimer's disease (AD) poses a significant healthcare challenge, with an escalating prevalence and a forecasted surge in affected individuals. The urgency for precise diagnostic tools to enable early interventions and improved patient care is evident. Despite advancements, existing detection frameworks exhibit limitations in accurately identifying AD, especially in its early stages. Model optimisation and accuracy are other issues. This paper aims to address this critical research gap by introducing ConvADD, an advanced Convolutional Neural Network architecture tailored for AD detection. By meticulously designing ConvADD, this study endeavours to surpass the limitations of current methodologies and enhance accuracy metrics, optimisation, and reliability of AD diagnosis. The dataset was collected from Kaggle and consists of preprocessed 2D images extracted from 3D images. Through rigorous experimentation, ConvADD demonstrates remarkable performance metrics, showcasing its potential as a robust and effective. The proposed model shows remarkable results with a tool for AD detection accuracy of 98.01%, precision of 98%, recall of 98%, and an F1-Score of 98%, with only 2.1 million parameters. However, despite its promising results, several challenges and limitations remain, such as generalizability across diverse populations and the need for further validation studies. By elucidating these gaps and challenges, this paper contributes to the ongoing discourse on improving AD detection methodologies and lays the groundwork for future research endeavours in this domain.

Author 1: Mohammed G Alsubaie

Author 2: Suhuai Luo

Author 3: Kamran Shaukat

Keywords: Alzheimer’s disease; AD detection; convolution neural network

Paper 32: A Cost-Effective IoT-based Transcutaneous Electrical Nerve Stimulation (TENS): Proof-of-Concept Design and Evaluation

Abstract: Transcutaneous electrical nerve stimulation (TENS) systems have been extensively used as a noninvasive and non-pharmaceutical approach for pain management and rehabilitation programs. Moreover, recent advances in telemedicine applications and the Internet of Things (IoT) have led to an increased interest in developing affordable systems that facilitate the remote monitoring of home-based therapeutic programs that help quantify usage and adherence, especially in clinical trials and research. Therefore, this study introduces the design and proof of concept validation of an IoT-enabled, cost-effective, single-channel TENS for remote monitoring of stimulation parameters. The presented prototype features programmable software that supports manipulating the stimulation parameters such as stimulation patterns, pulse width, and frequency. This flexibility can help researchers substantially investigate the effect of different stimulation parameters and develop subject-specific stimulation protocols. The IoT-based TENS system was built using commercial-grade electronic components controlled with open-source software. The system was validated for generating low-frequency (10 Hz) and high-frequency TENS stimulation (100 Hz). The developed system could produce constant biphasic pulses with an adjustable compliance voltage of 5–32 V. The stimulation current corresponding to the applied voltage was quantified across a resistive load of 1 kΩ, resulting in a stimulation current of approximately 4.88–28.79 mA. Furthermore, synchronizing the TENS system with an IoT platform provided the advantage of monitoring the usage and important stimulation parameters, which could greatly benefit healthcare providers. Hence, the proposed system discussed herein has the potential to be used in education, research, and clinics to investigate the effect of TENS devices in a variety of applications outside of the clinical setup.

Author 1: Ahmad O. Alokaily

Author 2: Meshael J. Almansour

Author 3: Ahmed A. Aldohbeyb

Author 4: Suhail S. Alshahrani

Keywords: Electro-stimulator; Internet of Things; TENS; pain management; smart health; IoT; telemedicine

Paper 33: An Intelligent Learning Approach for Improving ECG Signal Classification and Arrhythmia Analysis

Abstract: The development of deep learning algorithms in recent years has shown promise in interpreting ECGs, as these algorithms can be trained on large datasets and can learn to identify patterns associated with different heart conditions. The advantage of these algorithms is their ability to process large amounts of data quickly and accurately, which can help improve the speed and accuracy of diagnoses, especially for patients with heart conditions. Our proposed work provides performant models based on residual neural networks to automate the diagnosis of 12-lead ECG signals with more than 25 classes comprising different cardiovascular diseases (CVDs) and a healthy sinus rhythm. We conducted an experimental study using public datasets from Germany, the USA, and China and trained two models based on Residual Neural Net-works-50 (ResNet-50) and Xception from CNN techniques, which is one of the most effective classification models. Our models achieved high performances for both training and test tasks in terms of accuracy, precision, recall, and loss, with accuracy, recall, and precision exceeding 99.87% for the two proposed models during the training and validation. The loss obtained by the end of these two phases was 3.38.10-4. With these promising results, our suggested models can serve as diagnostic aids for cardiologists to evaluate ECG signals more quickly and objectively. Further quantitative and qualitative evaluations are presented and discussed in the study, and our work can be extended to other multi-modal big biological data tied with ECG for similar sets of patients to obtain a better understanding of the proposed approach for the benefit of the medical world.

Author 1: Sarah Allabun

Keywords: Electrocardiogram; cardiovascular diseases; classification; ResNet-50; Xception

Paper 34: Multi-Discriminator Image Restoration Algorithm Based on Hybrid Dilated Convolution Networks

Abstract: With the continuous development of generative adversarial networks (GAN), many image restoration problems that are difficult to solve based on traditional methods have been given new research avenues. Nevertheless, there are still problems such as structural distortion and texture blurring of the complemented image in the face of irregular missing. In order to overcome these problems and retrieve the lost critical data of the image, a two-stage image restoration complementation network is proposed in this paper. While introducing hybrid dilation convolution, two attention mechanisms are added to the network and optimized using multiple loss functions. This not only results in better image quality metrics, but also clearer and more coherent image details. In this paper, we tested the network on CelebA-HQ, Places2 and The Paris datasets and compared it with several classical image restoration models, such as GLC, Gconv, Musical and RFR, and the results proved that the complementary images in this paper are improved compared to the others.

Author 1: Chunming Wu

Author 2: Fengshuo Qi

Keywords: GAN; image restoration; hybrid dilated convolution; attention mechanism; two-stage network

Paper 35: Research on Resource Sharing Method of Library and Document Center Under the Multimedia Background

Abstract: In order to improve the utilization effect of the resources of the book and document center and ensure the security of its resource sharing, the resource sharing methods of the book and document center under the multimedia background are studied. The resource layer of this method is based on multimedia technology and combined with virtual technology to build a multimedia document cloud resource pool; At the same time, the adaptive clustering algorithm of empirical mode feature decomposition is used to obtain the number of document resources clustering and resource category labels, complete the resource clustering of the book and document center, and store it in the constructed resource pool; Users log in directly through the document resource sharing service of the service layer, and enter the resource center after authentication by the management layer. The service layer uses the regional document information resource co-construction and sharing mechanism based on blockchain to encrypt, co-identify and decrypt the clustered resources in the resource pool and then share the resources of the book and document center. The test results show that the clustering purity and contour coefficient of the method is above 0.970, and the clustering quality is good; The security of resource sharing is good, and the sensitivity result is 10.11% when the resource sharing ratio is 100%; It can effectively complete the resource sharing in the book and document center, and meet the sharing needs of book and document resources.

Author 1: Jianhui Zhang

Keywords: Multimedia background; library and reference center; resource sharing; virtual technology; multimedia technology

Paper 36: A Hybrid MCDM Model for Service Composition in Cloud Manufacturing using O-TOPSIS

Abstract: The purpose of this research article was to define the current or future usage of Industry 4.0 technologies (Cloud Computing, IoT, etc.) to improve industrial manufacturing. The goal of this study is to rate the options using a hybrid CRITIC - O-TOPSIS Multi Criteria Decision Making model. The CRITIC technique is used to calculate Objective Weights. Also, when comparing the findings to TOPSIS, A thorough Systematic Literature Review comes first. Secondly, a theoretical approach to recognizing the Index System of Criteria. Third, Creating a Hybrid Model of CRITIC and O-TOPSIS for Decision Making. Lastly, Comparing and Ordering Options. The proposed technique successfully addresses the ambiguity and uncertainty of heterogeneous information while maintaining assessment data accuracy. Also, because objective weights are more grounded in reality than subjective weights, the result is more precise. CRITIC approach results reveals that Ease of Opting has the most weight and Ease of Implementation has the least weight O-TOPSIS method ranks alternatives in the following order: A4>A5>A3>A1>A2. This paper ranks alternatives based on extensive 22 criteria in Service Composition in Cloud Manufacturing using the hybrid model CRITIC - O-TOPSIS

Author 1: Syed Omer Farooq Ahmed

Author 2: Adapa Gopi

Keywords: Cloud manufacturing (CMFg); CRITIC method; O-TOPSIS method; service composition

Paper 37: Comparative Analysis of Transformer Models for Sentiment Analysis in Low-Resource Languages

Abstract: The analysis of sentiments expressed on social media platforms is a crucial tool for understanding user opinions and preferences. The large amount of the texts found on social media are mostly in different languages. However, the accuracy of sentiment analysis in these systems faces different challenges in multilingual low-resource settings. Recent advancements in deep learning transformer models have demonstrated superior performance compared to traditional machine learning techniques. The majority of preceding works are predominantly constructed on the foundation of monolingual languages. This study presents a comparative analysis that assesses the effectiveness of transformer models, for multilingual low-resource languages sentiment analysis. The study aims to improve the accuracy of the existing baseline performance in analyzing tweets written in 12 low-resource African languages. Four widely used start-of-the-art transformer models were employed. The experiment was carried out using standard datasets of tweets. The study showcases AfriBERTa as a robust performer, exhibiting superior sentiment analysis capabilities across diverse linguistic contexts. It outperformed the established benchmarks in both SemEval-2023 Task 12 and AfriSenti baseline. Our framework achieves remarkable results with an F1-score of 81% and an accuracy rate of 80.9%. This study provides validation of the framework's robustness in the domain of sentiment analysis across a low-resource linguistics context. our research not only contributes a comprehensive sentiment analysis framework for low-resource African languages but also charts a roadmap for future enhancements. Emphasize the ongoing pursuit of adaptability and robustness in sentiment analysis models for diverse linguistic landscapes.

Author 1: Yusuf Aliyu

Author 2: Aliza Sarlan

Author 3: Kamaluddeen Usman Danyaro

Author 4: Abdulahi Sani B A Rahman

Keywords: Sentiment analysis; low-resource languages; multilingual; word-embedding; transformer

Paper 38: Influence of a Serious Video Game on the Behavior of Drivers in the Face of Automobile Incidents

Abstract: The primary objective of this research was to enhance driver behavior during incidents through the use of a serious video game. The study employed a true experimental design. The research population consisted of an unspecified number of drivers from the city of Trujillo. Sixty drivers from Trujillo were randomly selected, with 30 assigned to the control group and 30 to the experimental group. The experimental group utilized a video game developed in Unreal Engine 5.2.1., observation forms were used to gather information, and the collected data were subsequently analyzed and processed using the statistical software Jamovi v2.4.11. The results revealed a decrease equivalent to a 43.75% reduction in the number of action mistakes, a 51.14% reduction in the number of intention mistakes, a 31.4% decrease in the number of traffic law violations, and a 42.92% reduction in the number of aggressive attitudes. In conclusion, the use of a serious video game significantly improved driver behavior during incidents.

Author 1: Bryan S. Diaz-Sipiran

Author 2: Segundo E. Cieza-Mostacero

Keywords: Videogame; serious; behavior; driving; incidents

Paper 39: A Genetic Artificial Bee Colony Algorithm for Investigating Job Creation and Economic Enhancement in Medical Waste Recycling

Abstract: The effective management of end-of-life products, whether through recycling or incineration for electricity generation, holds pivotal significance amidst escalating concerns over economic, environmental, and social ramifications. While the economic and environmental dimensions often receive primary focus, the social aspect remains comparatively neglected within sustainability discourse. This paper undertakes a comprehensive exploration of the positive social impacts engendered by medical waste recycling, with a specific focus on job creation and economic value enhancement. The principal aim of this research is to highlight the social benefits derived from medical waste recycling, elucidating its role in fostering employment opportunities, and augmenting economic prosperity. By employing a Genetic Artificial Bee Colony algorithm, this study addresses two mathematical problems pertinent to optimizing recycling processes, thereby contributing to the advancement of sustainable waste management practices. Additionally, the proposed algorithm exhibits superior performance, highlighting its potential in addressing sustainability challenges. Ultimately, integrating the social dimension into end-of-life product management discussions can lead to a more comprehensive approach to sustainability, balancing environmental preservation with socio-economic progress.

Author 1: El Liazidi Sara

Author 2: Dkhissi Btissam

Keywords: Medical waste recycling; social impacts; genetic artificial bee colony algorithm; job creation; economic value

Paper 40: Multimodal Feature Fusion Video Description Model Integrating Attention Mechanisms and Contrastive Learning

Abstract: To avoid the issue of significant redundancy in the spatiotemporal features extracted from multimodal video description methods and the substantial semantic gaps between different modalities within video data. Building upon the TimeSformer model, this paper proposes a two-stage video description approach (Multimodal Feature Fusion Video Description Model Integrating Attention Mechanism and Contrastive Learning, MFFCL). The TimeSformer encoder extracts spatiotemporal attention features from the input video and performs feature selection. Contrastive learning is employed to establish semantic associations between the spatiotemporal attention features and textual descriptions. Finally, GPT2 is employed to generate descriptive text. Experimental validations on the MAVD, MSR-VTT, and VATEX datasets were conducted against several typical benchmark methods, including Swin-BERT and GIT. The results indicate that the proposed method achieves outstanding performance on metrics such as Bleu-4, METEOR, ROUGE-L, and CIDEr. The spatiotemporal attention features extracted by the model can fully express the video content and that the language model can generate complete video description text.

Author 1: Wang Zhihao

Author 2: Che Zhanbin

Keywords: Multimodal feature fusion; video description; spatiotemporal attention; comparative learning

Paper 41: Permanent Magnet Motor Control System Based on Fuzzy PID Control

Abstract: Although the traditional permanent magnet synchronous motor control system is simple and convenient, the control of speed and accuracy is often affected by external interference, which impacts the dynamic and static performance requirements. Therefore, this study attempts to introduce fuzzy rules to improve the proportional integral differential control method, and further integrate intelligent optimization algorithms into the fuzzy proportional integral differential control method to construct an efficient and feasible permanent magnet synchronous motor control method. The simulation experiment demonstrates that under fuzzy proportional integral differential control, there is no overshoot in the waveform when facing changes in load, and the tuning time increases from 0.01 seconds to 0.12 seconds. The steady-state error of speed control is small, and there is no obvious oscillation in the waveform. Fuzzy control enhances the control system. After the optimization of the artificial bee colony algorithm, the control system has a faster speed response, with the overshoot diminished from 11.2% to 3.1%, and the adjustment time reduced from 0.27 seconds to 0.19 seconds, enhancing its adaptability. Under load regulation, the optimized control system speed response curve responds in a timely manner without obvious overshooting and oscillating changes. Optimizing variable universe fuzzy proportional integral differential control enables the control system for having better static and dynamic performance, and enhances the adaptability and follow-up of the control system. The current curve starts to stabilize at 0.04s, overcoming the control system oscillations early. The speed response curve and the motor torque curve are improved by the optimized variable domain theory, and the amount of overshoot is significantly reduced. The research and design of a permanent magnet motor control system has practical significance for improving the application performance and adaptability of permanent magnet motors.

Author 1: Yin Sha

Author 2: Huwei Chen

Keywords: Permanent magnet motor; fuzzy PID; fuzzy control; automatic control system; artificial bee colony

Paper 42: Impact of Contradicting Subtle Emotion Cues on Large Language Models with Various Prompting Techniques

Abstract: The landscape of human-machine interaction is undergoing a transformation with the integration of conversational technologies. In various domains, Large Language Model (LLM) based chatbots are progressively taking on roles traditionally handled by human agents, such as task execution, answering queries, offering guidance, and delivering social and emotional assistance. Consequently, enhancing user satisfaction with these technologies is crucial for their effective incorporation. Emotions indeed play an effective role in responses generated by reinforcement-learning-based chatbots. In text-based prompts, emotions can be signaled by visual (emojis, emoticons) and linguistic (misspellings, tone of voice, word choice, sentence length, similes) aspects. Therefore, researchers are harnessing the power of Artificial Intelligence (AI) and Natural Language Processing techniques to imbue chatbots with emotional intelligence capabilities. This research aims to explore the impact of feeding contradicting emotional cues to the LLMs through different prompting techniques. The evaluation is based on specified instructions versus provided emotional signals. Each prompting technique is scrutinized by inducing a variety of emotions on widely used LLMs, ChatGPT 3.5 and Gemini. Instead of automating the prompting process, the prompts are given by exerting cognitive load to be more realistic regarding Human-Computer Interaction (HCI). The responses are evaluated using human-provided qualitative insights. The results indicate that simile-based cues have the highest impact in both ChatGPT and Gemini. However, results also conclude that the Gemini is more sensitive towards emotional cues. The finding of this research can benefit multiple fields: HCI, AI Development, Natural Language Processing, Prompt Engineering, Psychology, and Emotion analysis.

Author 1: Noor Ul Huda

Author 2: Sanam Fayaz Sahito

Author 3: Abdul Rehman Gilal

Author 4: Ahsanullah Abro

Author 5: Abdullah Alshanqiti

Author 6: Aeshah Alsughayyir

Author 7: Abdul Sattar Palli

Keywords: Emotion cues; prompt; Large Language Model (LLM); Human Computer Interactions (HCI)

Paper 43: Investigating Sampler Impact on AI Image Generation: A Case Study on Dogs Playing in the River

Abstract: AI image generation is a new and exciting field with many different uses. It is important to understand how different sampling techniques affect the quality of AI-generated images in order to get the best results. This study looks at how different sampling techniques affect the quality of AI-generated images of dogs playing in the river. This study is limited to a specific scenario, as there are not many images of dogs playing in the river already on the internet. The study used the Playground.ai open-source web platform to test different sampling techniques. DDIM was found to be the best sampling technique for generating realistic images of dogs playing in the river. Euler was also found to be very fast, which is an important consideration when choosing a sampling technique. These findings show that different sampling techniques have different strengths and weaknesses, and it is important to choose the right sampling technique for the specific task at hand. This study provides valuable insights into how sampling techniques affect AI image generation. It is important to choose the right sampling technique for the specific task at hand in order to get the best results. The study also demonstrates the societal relevance of AI-generated imagery in various applications.

Author 1: Sanjay Deshmukh

Keywords: Artificial Intelligence; image generation; filter; sampler; Euler; Heun

Paper 44: Enhancing Ultimate Bearing Capacity Assessment of Rock Foundations using a Hybrid Decision Tree Approach

Abstract: Accurately estimating the ultimate bearing capacity of piles embedded in rock is of paramount importance in the domains of civil engineering, construction, and foundation design. This research introduces an innovative solution to tackle this issue, leveraging a fusion of the Decision Tree method with two state-of-the-art optimization algorithms: the Zebra Optimization Algorithm and the Coronavirus Herd Immunity Optimizer. The research approach encompassed the creation of a hybridized model, unifying the DT with the Zebra Optimization Algorithm and Coronavirus Herd Immunity Optimizer. The primary objective was to augment the precision of the ultimate bearing capacity of prediction for piles embedded in rock. This hybridization strategy harnessed the capabilities of DT along with the two pioneering optimizers to address the inherent uncertainty stemming from diverse factors impacting bearing capacity. The Zebra Optimization Algorithm and Coronavirus Herd Immunity Optimizer showcased their efficacy in refining the base model, leading to substantial enhancements in predictive performance. This study's discoveries make a significant stride in the realm of geotechnical engineering by furnishing a sturdy approach to forecasting ultimate bearing capacity in rock-socketed piles. The hybridization method is a hopeful path for future research endeavors and practical implementations. Specifically, the DT + Zebra Optimization Algorithm model yielded dependable outcomes, as evidenced by their impressive R-squared value of 0.9981 and a low Root mean squared error value of 629.78. The attained outcomes empower engineers and designers to make well-informed choices concerning structural foundations in soft soil settings. Ultimately, this research advocates for safer and more efficient construction methodologies, mitigating the hazards linked to foundation failures.

Author 1: Mei Guo

Author 2: Ren-an Jiang

Keywords: Ultimate bearing capacity; decision tree; zebra optimization algorithm; coronavirus herd immunity optimizer

Paper 45: Improving Prediction Accuracy using Random Forest Algorithm

Abstract: One of the latest studies in predicting bankruptcy is the performance of the financial prediction models. Although several models have been developed, they often do not achieve high performance, especially when using an imbalanced data set. This highlights the need for more exact prediction models. This paper examines the application as well as the benefits of machine learning with the purpose of constructing prediction models in the field of corporate financial performance. There is a lack of scientific research related to the effects of using random forest algorithms in attribute selection and prediction process for enhancing financial prediction. This paper tests various feature selection methods along with different prediction models to fill the gap. The study used a quantitative approach to develop and propose a business failure model. The approach involved analyzing and preprocessing a large dataset of bankrupt and non-bankrupt enterprises. The performance of the model was then evaluated using various metrics such as accuracy, precision, and recall. Findings from the present study show that random forest is recommended as the best model to predict corporate bankruptcy. Moreover, findings write down that the proper use of attribute selection methods helps to enhance the prediction precision of the proposed models. The use of random forest algorithm in feature selection and prediction can produce more exact and more reliable results in predicting bankruptcy. The study proves the potential of machine learning techniques to enhance financial performance.

Author 1: Nesma Elsayed

Author 2: Sherif Abd Elaleem

Author 3: Mohamed Marie

Keywords: Corporate bankruptcy; feature selection; financial ratios; prediction models; random forest

Paper 46: StockBiLSTM: Utilizing an Efficient Deep Learning Approach for Forecasting Stock Market Time Series Data

Abstract: The article introduces a novel approach for forecasting stock market prices, employing a computationally efficient Bidirectional Long Short-Term Memory (BiLSTM) model enhanced with a global pooling mechanism. Based on the deep learning framework, this method leverages the temporal dynamics of stock data in both forward and reverse time frames, enabling enhanced predictive accuracy. Utilizing datasets from significant market players—HPQ, Bank of New York Mellon, and Pfizer—the authors demonstrate that the proposed single-layered BiLSTM model, optimized with RMSprop, significantly outperforms traditional Vanilla and Stacked LSTM models. The results are quantitatively evaluated using root mean squared error (RMSE), mean absolute error (MAE), and the coefficient of determination (R^2), where the BiLSTM model shows a consistent improvement in all metrics across different stock datasets. We optimized the hyperparameters tuning using two distinct optimizers (ADAM, RMSprop) on the HPQ, New York Bank, and Pfizer datasets. The dataset has been preprocessed to account for missing values, standardize the features, and separate it into training and testing sets. Moreover, line graphs and candlestick charts illustrate the models' ability to capture stock market trends. The proposed algorithms attained respective RMSE values of 0.413, 0.704, and 0.478. the proposed algorithms attained respective RMSE values of 0.413, 0.704, and 0.478. The results show the proposed methods' superiority over recently published models. In addition, it is concluded that the proposed single-layered BiLSTM-based architecture is computationally efficient and can be recommended for real-time applications involving Stock market time series data.

Author 1: Diaa Salama Abd Elminaam

Author 2: Asmaa M M. El-Tanany

Author 3: Mohamed Abd El Fattah

Author 4: Mustafa Abdul Salam

Keywords: Stock prediction; Univariate LSTM models; Deep learning; financial forecasting; Vanilla LSTM; Stacked LSTM; Bidirectional LSTM

Paper 47: Segmentation Analysis for Brain Stroke Diagnosis Based on Susceptibility-Weighted Imaging (SWI) using Machine Learning

Abstract: Magnetic Resonance Imaging (MRI) plays a crucial role in diagnosing brain disorders, with stroke being a significant category among them. Recent studies emphasize the importance of swift treatment for stroke, known as "time is brain," as early intervention within six hours of stroke onset can save lives and improve outcomes. However, the conventional manual diagnosis of brain stroke by neuroradiologists is subjective and time-consuming. To address this issue, this study presents an automatic technique for diagnosing and segmenting brain stroke from MRI images according to pre and post stroke patient. The technique utilizes machine learning methods, focusing on Susceptibility Weighted Imaging (SWI) sequences. The machine learning technique involves four stage, those are pre-processing, segmentation, feature extraction, and classification. In this paper, preprocessing and segmentation are proposed to identify the stroke region. The segmentation performance is assessed using Jaccard indices, Dice Coefficient, false positive, and false negative rates. The results show that adaptive threshold performs best for stroke lesion segmentation, with good improvement stroke patient that achieving the highest Dice coefficient of 0.96. In conclusion, this proposed stroke segmentation technique has promising potential for diagnosing early brain stroke, providing an efficient and automated approach to aid medical professionals in timely and accurate diagnoses.

Author 1: Shaarmila Kandaya

Author 2: Abdul Rahim Abdullah

Author 3: Norhashimah Mohd Saad

Author 4: Ezreen Farina

Author 5: Ahmad Sobri Muda

Keywords: Magnetic Resonance Imaging (MRI) diagnosis; time is brain; Susceptibility Weighted Imaging (SWI) and dice coefficient

Paper 48: A Genetic Algorithm-based Approach for Design-level Class Decomposition

Abstract: Software is always changed to accommodate environmental changes to preserve its existence. While changes happen to the software, the internal structure tends to decline in quality. The refactoring process is worth running to preserve the internal structure of the software. The decomposition process is a suitable refactoring process for Blob smell in class. It tried to split up the class based on the context in order to arrange it based on each responsibility. The previous approach has been implemented but still leaves problems. The optimum arrangement of class cannot be achieved using the previous approach. The genetic algorithm provides the search mechanism to find the optimum state based on the criterion stated at the beginning of the process. This paper presents the use of genetic algorithms to solve the design-level class decomposition problem. The paper explained several points, including the conversion from class to the chromosome construct, the fitness function calculation, selection, crossover, and mutation. The results show that the use of a genetic algorithm was able to solve the previous problems. The genetic algorithm can solve the local optimum problem from the previous approach. The increment of the fitness function of the study case proves it.

Author 1: Bayu Priyambadha

Author 2: Nobuya Takahashi

Author 3: Tetsuro Katayama

Keywords: Genetic algorithm; refactoring; class decomposition; blob smell; software internal quality

Paper 49: Analysis and Enhancement of Prediction of Cardiovascular Disease Diagnosis using Machine Learning Models SVM, SGD, and XGBoost

Abstract: Cardiovascular disease (CVD), claiming 17.9 million lives annually, is exacerbated by factors like high blood pressure and obesity, prompting extensive data collection for deeper insights. Machine learning aids in accurate diagnosis, with techniques like SVM, SGD, and XGBoost proposed for heart disease prediction, addressing challenges such as data imbalance and optimizing diagnostic accuracy. This study integrates these algorithms to improve cardiovascular disease diagnosis, aiming to reduce mortality rates through timely interventions. This research investigates the efficacy of Support Vector Machine (SVM), Stochastic Gradient Descent (SGD), and XGBoost machine learning techniques for heart disease prediction. Analysis of the models' performance metrics reveals distinct characteristics and capabilities. SVM demonstrates robust performance with a training accuracy of 88.28% and a model accuracy score of 87.5%, exhibiting high precision and recall values across both classes. SGD, while commendable with a training accuracy of 83.65% and a model accuracy score of 84.24%, falls slightly behind SVM in accuracy and precision. XGBoost Classifier showcases perfect training accuracy but potential overfitting, yet demonstrates comparable precision and recall values to SVM. Overall, SVM emerges as the most effective model for heart disease prediction, followed by SGD and XGBoost Classifier. Further optimization and investigation into generalization capabilities are recommended to enhance the performance of SGD and XGBoost Classifier in clinical settings.

Author 1: Sandeep Tomar

Author 2: Deepak Dembla

Author 3: Yogesh Chaba

Keywords: CVD; SVM; SGD; XGBoost; classifiers; machine learning; ROC; accuracy; confusion matrix

Paper 50: Towards a New Artificial Intelligence-based Framework for Teachers’ Online Continuous Professional Development Programs: Systematic Review

Abstract: In recent years, the Artificial Intelligence (AI) field has witnessed rapid growth, affecting diverse sectors, including education. In this systematic review of literature, we aimed to analyze studies concerning the integration of AI in the continuous professional development (CPD) of teachers in order to generate a global vision on its potential to enhance the quality of CPD programs in the international level, and to provide recommendations for its application in the Moroccan context. To achieve our objective, we conducted a research that involves a review of international indexed databases (Scopus, Web of Science, Eric) published between 2019 and 2023 using PICO framework to formulate our search query and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework to select 25 relevant studies based on include and exclude criteria like publishing year, type of documents, publishing mode, subject area, language, and other criteria. The results reveal that AI integration has a positive impact on CPD programs by offering beneficial intelligent tools that can tailor adaptive training programs to meet teachers’ specific needs, preferences, and proficiency levels. Furthermore, our findings identify the importance of integrating AI as a core topic within CPD programs to enhance teachers’ AI literacy, enabling them to effectively navigate and utilize AI-based tools in their educational environment. This is important for preparing teachers to engage with the technological advances shaping the educational system. In conclusion, our systematic review emphasizes the significance of AI integration in CPD programs and offers tailored recommendations for its implementation in the Moroccan educational context. By adopting these recommendations, Morocco will pave the way for a dynamic CPD framework that meets the evolving needs of educators and students alike.

Author 1: Hamza Fakhar

Author 2: Mohammed Lamrabet

Author 3: Noureddine Echantoufi

Author 4: Khalid El khattabi

Author 5: Lotfi Ajana

Keywords: Artificial intelligent; continuous professional development; Moroccan in-service teacher; digital teacher; online training; adaptive development

Paper 51: Improvement of Social Skills in Children with Autism Spectrum Disorder Through the use of a Video Game

Abstract: The main research objective was to improve social skills through a video game, the type of research was applied with a pure experimental design, with a sample of 60 children with autism spectrum disorder from the Christa McAuliffe school, randomly allocated 30 to the control group (CG) and 30 to the experimental group (GE), the latter using a video game developed with the Unity 3D; Data collection was carried out by means of an adapted test from the cited authors; subsequently, the data collected was analyzed and processed using the Jamovi v2 statistical software. 3.28. The results obtained were an increase of 27.8% on average in the level of communication skills, an increase of 22.4% on average in the level of skills related to feelings, an increase of 20.4% on average in the level of skills alternative to violence and an increase of 19% on average in the level of Pro-amical skills. It was concluded that, the use of a video game significantly improved social skills.

Author 1: Luis C. Soles-Núñez

Author 2: Segundo E. Cieza-Mostacero

Keywords: Video games; social skills; autism spectrum disorder; SUM methodology; academic software

Paper 52: Cinematic Curator: A Machine Learning Approach to Personalized Movie Recommendations

Abstract: This work suggests a sophisticated movie recommendation system that offers individualized recommendations based on user preferences by combining content-based filtering, collaborative filtering, and deep learning approaches. The system use natural language processing (NLP) to examine user-generated content, movie summaries, and reviews in order to get a sophisticated comprehension of thematic aspects and narrative styles. The model includes SHAP for explainability to improve transparency and give consumers insight into the reasoning behind recommendations. The user-friendly interface, which is accessible via web and mobile applications, guarantees a smooth experience. The system is able to adjust to changing user preferences and market trends through ongoing upgrades that are founded on fresh data. The system's efficacy is validated by user research and A/B testing, which show precise and customized movie recommendations that satisfy a range of tastes.

Author 1: B. Venkateswarlu

Author 2: N. Yaswanth

Author 3: A. Manoj Kumar

Author 4: U. Satish

Author 5: K. Dwijesh

Author 6: N. Sunanda

Keywords: Machine learning algorithms decision tree; random forest model-evaluation; accuracy value; precision value; F1 score

Paper 53: Sentiment Analysis of Pandemic Tweets with COVID-19 as a Prototype

Abstract: One of the most important applications of text mining is sentiment analysis of pandemic tweets. For example, it can make governments able to predict the onset of pandemics and to put in place safe policies based on people's feelings. Many research studies addressed this issue using various datasets and models. Nevertheless, this is still an open area of research in which many datasets and models are yet to be explored. This paper is interested in the sentiment analysis of COVID-19 tweets as a prototype. Our literature review revealed that as the dataset size increases, the accuracy generally tends to decrease. This suggests that using a small dataset might provide misleading results that cannot be generalized. Hence, it is better to consider large datasets and try to improve analysis performance on it. Accordingly, in this paper we consider a huge dataset, namely COVIDSenti, which is composed of three sub datasets (COVIDSenti_A, COVIDSenti_B, and COVIDSenti_C). These datasets have been processed with a number of Machine Learning (ML) models, Deep Learning (DL) models, and transformers. In this paper, we examine other ML and DL models aiming to find superior solutions. Specifically, we consider Ridge Classifier (RC), Multinomial Naïve Bayes (MNB), Stochastic Gradient Descent (SGD), Support Vector Classification (SVC), Extreme Gradient Boosting (XGBoost), and the DL Gated Recurrent Unit (GRU). Experimental results have shown that unlike the models that we tested, and the state-of-the-art models on the same dataset, SGD technique with count vectorizer showed quite constantly high performance on all the four datasets.

Author 1: Mashail Almutiri

Author 2: Mona Alghamdi

Author 3: Hanan Elazhary

Keywords: COVID-19; deep learning; machine learning; sentiment analysis; text mining; tweets

Paper 54: Automating Mushroom Culture Classification: A Machine Learning Approach

Abstract: Traditionally, the classification of mushroom cultures has conventionally relied on manual inspection by human experts. However, this methodology is susceptible to human bias and errors, primarily due to its dependency on individual judgments. To overcome these limitations, we introduce an innovative approach that harnesses machine learning methodologies to automate the classification of mushroom cultures. Our methodology employs two distinct strategies: the first involves utilizing the histogram profile of the HSV color space, while the second employs a convolutional neural network (CNN)-based technique. We evaluated a dataset of 1400 images from two strains of Pleurotus ostreatus mycelium samples over a period of 14 days. During the cultivation phase, we base our operations on the histogram profiles of the masked areas. The application of the HSV histogram profile led to an average precision of 74.6% for phase 2, with phase 3 yielding a higher precision of 95.2%. For CNN-based method, the discriminative image features are extracted from captured images of rhizomorph mycelium growth. These features are then used to train a machine learning model that can accurately estimate the growth rate of a rhizomorph mycelium culture and predict contamination status. Using MNet and MConNet approach, our results achieved an average accuracy of 92.15% for growth prediction and 97.81% for contamination prediction. Our results suggest that computer-based approaches could revolutionize the mushroom cultivation industry by making it more efficient and productive. Our approach is less prone to human error than manual inspection, and it can be used to produce mushrooms more efficiently and with higher quality.

Author 1: Hamimah Ujir

Author 2: Irwandi Hipiny

Author 3: Mohamad Hasnul Bolhassan

Author 4: Ku Nurul Fazira Ku Azir

Author 5: SA Ali

Keywords: Machine learning; convolution neural networks; mushroom cultivation; rhizomorph mycelium

Paper 55: Leather Image Quality Classification and Defect Detection System using Mask Region-based Convolution Neural Network Model

Abstract: The leather industry is increasingly becoming one amongst the most important manufacturing industries in the world. Increasing demand has posed a great challenge as well as an opportunity for these industries. Quality of a leather product has been always the main factor in the setting of the market selling price. Usually, quality control is done with manual inspection. However, with human related errors such as fatigue, loss of concentration, etc., misclassification of the produced leather quality becomes a very serious issue. To tackle this issue, traditionally, image processing algorithms have been used, but, have not been effective due to low accuracies and high processing time. The introduction of Deep Learning methodologies such as Convolutional Neural Networks (CNNs), however, makes image classification much simpler. It incorporates automated feature learning and extraction, giving accurate results in lesser time. In addition, the usage of deep learning can also be applied for defect detection, which is, locating defects in the image. In this paper, a system for leather image classification and defect detection is proposed. Initially, the captured images are sent to a classification system, which classifies the image as good quality or defect quality. If the output of the classification system is defect quality, then a defect detection system works on the images, and locates the defects in the image. The classification system and the defect detection system are developed using Inception V3 CNN and Mask R-CNN respectively. Experimental results using these CNNs have shown great potential with respect to object classification and detection, which, with further development can give unparalleled performance for applications in these fields.

Author 1: Azween Bin Abdullah

Author 2: Malathy Jawahar

Author 3: Nalini Manogaran

Author 4: Geetha Subbiah

Author 5: Koteeswaran Seeranagan

Author 6: Balamurugan Balusamy

Author 7: Abhishek Chengam Saravanan

Keywords: Image leather classification; leather defect detection; Convolutional Neural Network; CNN; deep learning

Paper 56: Prediction of Pigment Epithelial Detachment in Optical Coherence Tomography Images using Machine Learning

Abstract: Pigment Epithelial Detachment (PED) is an eye condition that can affect adults over 50 and eventually harm their central vision. The PED region is positioned between the Bruch's membrane (BM) and the RPE (Retinal Pigment Epithelium) layer. Due to PED, the RPE layer is elevated arc shaped. In this paper, a method to extract the best features to detect pigment epithelial detachment (PED) is proposed. This method uses four-stage strategy that drew inspiration from OCT (Optical Coherence Tomography) imaging to detect the PED. In the first stage, to reduce the speckle-noise, in the second stage, segment the Retinal Pigment Epithelium (RPE) layer. In the third stage, a novel method is proposed to extract the best features to detect PED, and in the fourth stage, machine learning classifiers such as K-Nearest Neighbors (KNN), Logistic Regression (LR), Naïve ayes (NB), and Artificial Neural Networks (ANN) were used to significantly predict the PED. For experimental results, 150 retinal OCT volumes were used, 75 normal OCT volumes, and 75 pigment epithelial detachment volumes. Among the 150 images, 80% were used for training and 20% were used for testing. Here, there are 30 images for testing and 120 images for training. To generate a confusion matrix based on the matrices are true positive (TP), false positive (FP), true negative (TN), and false negative (FN). Logistic Regression is predicted more accuracy among the ANN, LR, NB, and KNN models. The LR model predicted accuracy 96.67% for PED detection.

Author 1: T. M. Sheeba

Author 2: S. Albert Antony Raj

Keywords: Artificial neural network; k-nearest neighbor; logistic regression; layer segmentation; naïve base; optical coherence tomography; pigment epithelial detachment

Paper 57: Investigating the Effect of Small Sample Process Capability Index Under Different Bootstrap Methods

Abstract: In the quality control of multi-variety and small-batch products, the calculation of the process capability index is particularly important. However, when the sample size is not enough, the process distribution cannot be judged, if the traditional method is still used to calculate the process capability index; there will be misapplication or misuse. In this paper, the Bootstrap method is introduced into the estimation of process capability index and the calculation of its confidence interval by using Standard Bootstrap (SB), Percentile Bootstrap (PB), Percentile-t Bootstrap (PTB) and Biased-corrected Percentile Bootstrap (BCPB) methods were used to analyze and compare the process capability index. It is found that in symmetric distribution, only the sample size has a significant effect on the length of the confidence interval；but in asymmetric distribution, sample size and Bootstrap methods are both significant factors affecting the length of confidence interval.

Author 1: Liyan Wang

Author 2: Guihua Bo

Author 3: Mingjuan Du

Keywords: Process capability indices; bootstrap; confidence interval; small samples

Paper 58: Discovering the Global Landscape of Agri-Food and Blockchain: A Bibliometric Review

Abstract: The agri-food supply chain encompasses all the entities involved in the production and processing of food, from producers to consumers. Traceability is crucial in ensuring that food products are available, affordable, and accessible. Blockchain technology has been proposed as a way to improve traceability in the agri-food supply chain by providing transparency and trust. However, research in this area is still in its early stages. This study aims to examine the trend of blockchain in agri-food supply chain traceability for food security. A bibliometric analysis was conducted on 1047 scholarly works from the Scopus database, starting in 2016. The analysis looked at citation patterns and the development of blockchain technology in agri-food supply chain research and identified trends by source title, nation, institution, and key players. The analysis also examined the frequency of keywords, titles, and abstracts to identify key themes. The analysis has revealed a strong correlation between blockchain technology and traceability in the agri-food supply chain, indicating a promising area for further research. The results show that blockchain-based research for traceability in the agri-food supply chain has increased and is being widely distributed, particularly in regions beyond Europe. The potential benefits it can bring to the supply chain will contribute to the success of the Sustainable Development Goals (SDGs) by ensuring a safe and sufficient global food supply.

Author 1: Sharifah Khairun Nisa’ Habib Elias

Author 2: Sahnius Usman

Author 3: Suriayati Chuprat

Keywords: Agri-food supply chain; bibliometric; blockchain; traceability

Paper 59: A Robust Hybrid Convolutional Network for Tumor Classification Using Brain MRI Image Datasets

Abstract: Brain tumour detection is challenging for experts or doctors in the early stage. Many advanced techniques are used for the detection of different cancers and analysis using different medical images. Deep learning (DL) comes under artificial intelligence, which is used to analyse and characterisation medical image processing and also finds the classification of brain cancer. Magnetic Resonance Imaging (MRI) has become the keystone in brain cancer recognition and the fusion of advanced imaging methods with cutting-edge DL models has exposed great potential in enhancing accuracy. This research aims to develop an efficient hybrid CNN model by employing support vector machine (SVM) classifiers to advance the efficacy and stability of the projected convolutional neural network (CNN) model. Two distinct brain MRI image datasets (Dataset_MC and Dataset_BC) are binary and multi-classified using the suggested CNN and hybrid CNN-SVM (Support Vector Machine) models. The suggested CNN model employs fewer layers and parameters for feature extraction, while SVM functions as a classifier to preserve maximum accuracy in a shorter amount of time. The experiment result shows the evaluation of the projected CNN model with the SVM for the performance evaluation, in which CNN-SVM give the maximum accuracy on the test datasets at 99% (Dataset_BC) and 98% (Dataset_MC) as compared to other CNN models.

Author 1: Satish Bansal

Author 2: Rakesh S Jadon

Author 3: Sanjay K. Gupta

Keywords: CNN; SVM; MRI images; brain tumor; deep learning

Paper 60: Emotion Recognition with Intensity Level from Bangla Speech using Feature Transformation and Cascaded Deep Learning Model

Abstract: Speech Emotion Recognition (SER) identifies and categorizes emotional states by analyzing speech signals. The intensity of specific emotional expressions (e.g., anger) conveys critical directives and plays a crucial role in social behavior. SER is intrinsically language-specific; this study investigated a novel cascaded deep learning (DL) model to Bangla SER with intensity level. The proposed method employs the Mel-Frequency Cepstral Coefficient, Short-Time Fourier Transform (STFT), and Chroma STFT signal transformation techniques; the respective trans-formed features are blended into a 3D form and used as the input of the DL model. The cascaded model performs the task in two stages: classify emotion in Stage 1 and then measure the intensity in Stage 2. DL architecture used in both stages is the same, which consists of a 3D Convolutional Neural Network (CNN), a Time Distribution Flatten (TDF) layer, a Long Short-term Memory (LSTM), and a Bidirectional LSTM (Bi-LSTM). CNN first extracts features from 3D formed input; the features are passed through the TDF layer, Bi-LSTM, and LSTM; finally, the model classifies emotion along with its intensity level. The proposed model has been evaluated rigorously using developed KBES and other datasets. The proposed model revealed as the best-suited SER method compared to existing prominent methods achieving accuracy of 88.30% and 71.67% for RAVDESS and KBES datasets, respectively.

Author 1: Md. Masum Billah

Author 2: Md. Likhon Sarker

Author 3: M. A. H. Akhand

Author 4: Md Abdus Samad Kamal

Keywords: Bangla speech emotion recognition; speech signal transformation; convolutional neural network; bidirectional long short-term memory

Paper 61: Optimizing Deep Learning for Efficient and Noise-Robust License Plate Detection and Recognition

Abstract: Accurate license plate recognition (LPR) remains a crucial task in various applications, from traffic monitoring to security systems. However, noisy environments with challenging factors like blurred images, low light, and complex backgrounds can significantly impede traditional LPR methods. This work proposes a deep learning based LPR system optimized for performance in noisy environments through hyperparameter tuning and bounding box refinement. We first preprocessed the noisy images by noise reduction which is crucial for robust LPR. We employed Convolutional Autoencoder (CAE) trained on noisy/clean image pairs to remove noise and enhance details. We utilized the InceptionResNetV2 architecture, pre-trained on ImageNet, for its strong feature extraction capabilities. We then added Region Proposal Network (RPN) head added to InceptionResNetV2 to predict candidate bounding boxes around potential license plates. We employed grid search to optimize key hyperparameters like learning rate, optimizer settings, and RPN anchor scales, ensuring optimal model performance for the specific noise patterns in the target dataset. Non-maximum suppression (NMS) eliminates redundant proposals, and a separate detection head classifies each remaining bounding box as license plate or background. Finally, bounding boxes are refined for improved accuracy. For confirmed license plates, a Bidirectional LSTM/CRNN network extracts and decodes character sequences within the refined bounding boxes. Compared to the recent methods, the proposed approach yielded the highest detection and recognition performance in noisy environments which can be best utilized for monitoring traffic, security systems in noisy environment. Our optimized LPR system demonstrates significantly improved accuracy and robustness compared to baseline methods, particularly in noisy environments.

Author 1: Seong-O Shim

Author 2: Romil Imtiaz

Author 3: Safa Habibullah

Author 4: Abdulrahman A. Alshdadi

Keywords: De-noising; image analysis; image processing; computer vision; image restoration

Paper 62: Crowdsourcing Requirements Engineering: A Taxonomy-based Review

Abstract: Interesting insights have been found by the research community indicating that early user involvement in Requirements Engineering (RE) has a considerable association with higher requirements quality, software project success and as well boosting user loyalty. In addition, traditional RE approaches confront scalability issues and would be time consuming and expensive to be applied with contemporary applications that can be surrounded by a large crowd. Therefore, recent attention has been shed on leveraging the principle of Crowdsourcing (CS) in requirements engineering. Engaging the crowd in RE activities has been researched by several studies. Hence, we synthesize and review the literature of the knowledge domain Crowdsourcing Requirements Engineering using a proposed taxonomy of the area. A total of 52 studies were selected for review in this paper. The review aims to provide the potential directions in the area and pave the way for other researchers to understand it and find possible gaps.

Author 1: Ghadah Alamer

Author 2: Sultan Alyahya

Author 3: Hmood Al-Dossari

Keywords: Crowdsourcing requirements engineering; crowdsourcing; CrowdRE; crowd

Paper 63: An Effective Book Recommendation System using Weighted Alternating Least Square (WALS) Approach

Abstract: Book recommendation systems are essential resources for connecting people with the correct books, encouraging a love of reading, and sustaining a vibrant literary ecosystem in an era when information overload is a prevalent problem. With the emergence of digital libraries and large online book retailers, readers may no longer find their next great literary journey without the help of customized book suggestions. This work offers a novel way to improve book recommendation systems using the Weighted Alternating Least Squares (WALS) technique, which is intended to uncover meaningful patterns in user ratings. The suggested approach minimizes the Root Mean Square Error (RMSE), a crucial indicator of recommendation system (RS) performance, in order to tackle the problem of optimizing recommendations. By representing user-item interactions as a matrix factorization problem, the WALS approach improves the recommendation process. In contrast to conventional techniques, WALS adds weighted elements that highlight specific user-item pairings' significance, increasing the recommendations' accuracy. Through an empirical study, the proposed approach demonstrates a significant reduction in RMSE when compared to standard RS, highlighting its effectiveness in enhancing the quality of book recommendations. By leveraging weighted matrix factorization, the proposed method adapts to the nuanced preferences and behaviors of users, resulting in more accurate and personalized book recommendations. This advancement in recommendation technology is poised to benefit both readers and the book industry by fostering more engaging and satisfying reading experiences.

Author 1: Kavitha V K

Author 2: Sankar Murugesan

Keywords: Recommendation system; user ratings; matrix factorization; alternating least square; weighted matrix factorization

Paper 64: Improving Predictive Maintenance in Industrial Environments via IIoT and Machine Learning

Abstract: Optimizing maintenance procedures is essential in today's industrial settings to reduce downtime and increase operational effectiveness. To improve predictive maintenance in industrial settings, this article investigates the combination of machine learning (ML) techniques and the Industrial Internet of Things (IIoT). The goal of this research is to advance predictive maintenance in industrial settings by integrating ML with IIoT in a seamless manner. Addressing the complexities of industrial systems and limitations of traditional maintenance methods, this study presents a methodology leveraging four distinct ML models. The technique includes a thorough assessment of these models' correctness, revealing differences that highlight the significance of a careful model selection procedure. The current investigation analysis finds the most effective model for predictive maintenance activities using thorough data analysis and visualization. Our work offers a potential path forward for the industrial sector and provides insights into the complex interactions between IIoT and ML. This study lays the groundwork for future developments in predictive maintenance, which will reduce downtime and extend the life of industrial equipment.

Author 1: Saleh Othman Alhuqayl

Author 2: Abdulaziz Turki Alenazi

Author 3: Hamad Abdulaziz Alabduljabbar

Author 4: Mohd Anul Haq

Keywords: Predictive maintenance; IIoT; data visualization; machine learning; industrial systems

Paper 65: Analyzing Privacy Implications and Security Vulnerabilities in Single Sign-On Systems: A Case Study on OpenID Connect

Abstract: Single Sign-On (SSO) systems have gained popularity for simplifying the login process, enabling users to authenticate through a single identity provider (IDP). However, their widespread adoption raises concerns regarding user privacy, as IDPs like Google or Facebook can accumulate extensive data on user web behavior. This presents a significant challenge for privacy-conscious users seeking to restrict disclosure of their online activities to third-party entities. This paper presents a comprehensive study focused on the OpenID Connect protocol, a widely utilized SSO standard. Our analysis delves into the protocol's operation, identifying security flaws and vulnerabilities across its various stages. Additionally, we systematically examine the privacy implications associated with user access to SSO systems. We offer a detailed account of how easily user information can be accessed, shedding light on potential risks. The findings underscore the imperative to address privacy vulnerabilities within SSO infrastructures. We advocate for proactive measures to enhance system security and safeguard user privacy effectively. By identifying weaknesses in the OpenID Connect protocol and its implementations, stakeholders can implement targeted strategies to mitigate risks and ensure the protection of user data. This research aims to foster a more secure and privacy-respecting environment within the evolving landscape of SSO systems.

Author 1: Mohammed Al Shabi

Author 2: Rashiq Rafiq Marie

Keywords: Single Sign-On; OpenID connect protocol; vulnerabilities; privacy; third-party

Paper 66: A Patrol Platform Based on Unmanned Aerial Vehicle for Urban Safety and Intelligent Social Governance

Abstract: Urban patrols can detect emergencies in a timely manner and collect information, which helps to improve the quality of services in the city and enhance the comfort of residents. This study proposes the use of IoT-based drones for urban patrol tasks, aiming to explore the potential applications of drones in smart city governance. The main technical challenge in the process of urban patrols by drones is how to plan a flight path for them. Therefore, this article first designs a smart patrol system based on drones and Internet of Things (IoT). Meanwhile, as information collection is an important aspect of urban patrol tasks, a mathematical model with the goal of maximizing information collection has been established to provide cost-effective patrol services. On this basis, in order to improve the accuracy of crow search algorithm (CSA), differential crow search strategy and variable flight step size are designed. In addition, the Levy flight strategy is introduced into the traditional CSA algorithm, and an improved crow search algorithm (ICSA) is proposed. Finally, a corresponding simulation environment was established based on the actual urban scene and compared with other algorithms. The numerical results indicate that compared with the other three swarm intelligence algorithms, the algorithm designed in this paper has more superiority.

Author 1: Ying Yang

Author 2: Rui Ma

Author 3: Fengjiao Zhou

Keywords: Patrol drones; trajectory planning; smart city governance; crow search algorithm; swarm intelligence algorithm

Paper 67: Entity Relation Joint Extraction Method Based on Insertion Transformers

Abstract: Existing multi-module multi-step and multi-module single-step methods for entity relation joint extraction suffer from issues such as cascading errors and redundant mistakes. In contrast, the single-module single-step modeling approach effectively alleviates these limitations. However, the single-module single-step method still faces challenges when dealing with complex relation extraction tasks, such as excessive negative samples and long decoding times. To address these issues, this paper proposes an entity relation joint extraction method based on Insertion Transformers, which adopts the single-module single-step approach and integrates the newly proposed tagging strategy. This method iteratively identifies and inserts tags in the text, and then effectively reduces decoding time and the count of negative samples by leveraging attention mechanisms combined with contextual information, while also resolving the problem of entity overlap. Compared to the state-of-the-art models on two public datasets, this method achieves high F1 scores of 93.2% and 91.5%, respectively, demonstrating its efficiency in resolving entity overlap issues.

Author 1: Haotian Qi

Author 2: Weiguang Liu

Author 3: Fenghua Liu

Author 4: Weigang Zhu

Author 5: Fangfang Shan

Keywords: Entity relation extraction; tagging strategy; joint extraction; transformer

Paper 68: Timber Defect Identification: Enhanced Classification with Residual Networks

Abstract: This study investigates the potential enhancement of classification accuracy in timber defect identification through the utilization of deep learning, specifically residual networks. By exploring the refinement of these networks via increased depth and multi-level feature incorporation, the goal is to develop a framework capable of distinguishing various defect classes. A sequence of ablation experiments was conducted, comparing our proposed framework's performance (R1, R2 and R3) with the original ResNet50 architecture. Furthermore, the framework’s classification accuracy was evaluated across different timber species and statistical analyses such as independent t-tests and one-way ANOVA tests were conducted to identify the significant differences. Results showed that while the R1 architecture demonstrated slight improvement over ResNet50, particularly with the addition of an extra layer ("ConvG"), the latter still maintained superior overall performance in defect identification. Similarly, the R2 architecture, despite achieving notable accuracy improvements, slightly lagged behind R1. Integration of fully pre-activation activation functions in the R3 architecture yielded significant enhancements, with a 14.18% increase in classification accuracy compared to ResNet50. The R3 architecture showcased commendable defect identification performance across various timber species, though with slightly lower accuracy in Rubberwood. Nonetheless, its performance surpassed both ResNet50 and other proposed architectures, suggesting its suitability for timber defect identification. Statistical analysis confirmed the superiority of the R3 architecture across multiple timber species and this underscores the significance of integrating network depth and fully pre-activation activation functions in improving classification performance. In conclusion, while the wood industry has made strides towards automation in timber grading, significant challenges remain. Overcoming these challenges will require innovative approaches and advancements in image processing and artificial intelligence to realize the full potential of automated grading systems.

Author 1: Teo Hong Chun

Author 2: Ummi Raba’ah Hashim

Author 3: Sabrina Ahmad

Keywords: Residual neural network; convolutional neural network; timber defect identification; deep learning

Paper 69: Enhancing Supply Chain Management Efficiency: A Data-Driven Approach using Predictive Analytics and Machine Learning Algorithms

Abstract: Contemporary firms rely heavily on the effectiveness of their supply chain management. Modern supply chains are complicated and unpredictable, and traditional methods frequently find it difficult to adjust to these factors. Increasing supply chain efficiency through improved supplier performance, demand prediction, inventory optimisation, and streamlined logistics processes may be achieved by utilising sophisticated data analytics and machine learning approaches. In order to improve supply chain management efficiency, this study suggests a unique data-driven strategy that makes use of Deep Q-Learning (DQL). The goal is to create optimisation frameworks and prediction models that can support well-informed decision-making and supply chain operational excellence. The deep Q learning technique is thoroughly integrated into supply chain management in this study, which makes it innovative. The suggested framework gives a comprehensive method for tackling the difficulties of contemporary supply chain management by integrating cutting-edge methodologies including demand forecasting, inventory optimisation, supplier performance prediction, and logistics optimisation. Predictive modelling, performance assessment, and data preparation are three of the proposed framework's essential elements. Cleansing and converting raw data to make it easier to analyse is known as data preparation. To create machine learning frameworks for applications like demand forecasting and logistics optimization, predictive modelling uses DQL. The method's efficacy in raising supply chain efficiency is evaluated through performance evaluation and acquired 98.9% accuracy while implementation. Findings show that the suggested DQL-based strategy is beneficial. Demand is precisely predicted using predictive models, which improves inventory control and lowers stockouts. Supply chain efficiencies brought about by DQL-based optimisation algorithms include lower costs and better service quality. Performance assessment measures show notable gains above baseline methods, highlighting the importance of DQL in supply chain management. This study demonstrates how Deep Q-Learning has the ability to completely change supply chain management procedures. In today's dynamic environment, organisations may gain competitive advantage and sustainable development through supply chain operations that are more efficient, agile, and resilient thanks to the incorporation of modern analytical methodologies and data-driven insights.

Author 1: Shamrao Parashram Ghodake

Author 2: Vinod Ramchandra Malkar

Author 3: Kathari Santosh

Author 4: L. Jabasheela

Author 5: Shokhjakhon Abdufattokhov

Author 6: Adapa Gopi

Keywords: Supply chain management; predictive analytics; demand forecasting; inventory management; exploratory data analysis

Paper 70: Advancing Prostate Cancer Diagnostics with Image Masking Techniques in Medical Image Analysis

Abstract: Prostate cancer is a prevalent health concern characterized by the abnormal and uncontrolled growth of cells within the prostate gland in men. This research paper outlines a standardized methodology for integrating medical slide images into machine learning algorithms, specifically emphasizing advancing healthcare diagnostics. The methodology involves thorough data collection, exploration, and image analysis, establishing a foundation for future progress in medical image analysis. The study investigates the relationships among image characteristics, data providers, and target variables to reveal patterns conducive to diagnosing medical conditions. Novel background prediction techniques are introduced, highlighting the importance of meticulous data preparation for improved diagnostic accuracy. The results of our research offer insights into dataset characteristics and image dimensions, facilitating the development of machine-learning models for healthcare diagnosis. Through deep learning and statistical analysis, we contribute to the evolving field of prostate cancer detection, showcasing the potential of advanced imaging modalities. This research promises to revolutionize healthcare diagnostics and shape the trajectory of medical image analysis, providing a robust framework for applying machine learning algorithms in the field. The standardized approach presented in this paper aims to enhance the reproducibility and comparability of studies in medical image analysis, fostering advancements in healthcare technology.

Author 1: H. V. Ramana Rao

Author 2: V RaviSankar

Keywords: Prostate cancer; data exploration; image analysis; medical conditions; background prediction techniques; data preparation; diagnostic accuracy; dataset characteristics; image dimensions; deep learning; statistical analysis; prostate cancer detection; advanced imaging modalities; healthcare diagnostics; medical image analysis; machine learning; target variables

Paper 71: Deep Learning Network Optimization for Analysis and Classification of High Band Images

Abstract: Examination and categorization of high-band pictures are used to describe the process of analysing and classifying photos that have been taken in many bands. Deep learning networks are known for their capacity to extract intricate information from images with a high bandwidth. The novelty lies in the integration of adaptive motion optimization, spectral-spatial transformer for categorization, and CNN-based feature extraction, enhancing high-band picture search efficiency and accuracy. The three primary parts of the technique are adaptive motion for optimization, spectral-spatial transformer for categorization, and CNN-based feature extraction. Initially, hierarchical characteristics from high-band pictures using a CNN. The CNN method enables precise feature representation and does a good job of matching the image's high and low features. This transformer module modifies the spectral and spatial properties of pictures intended for usage, enabling more careful categorization. This method performs better when processing complicated and variable picture data by integrating spectral and spatial information. Additionally, it is preferable to incorporate adaptive motion algorithms into offering the deep learning network training set. During training, this optimization technique dynamically modifies the motion parameter for quicker convergence and better generalization performance. The usefulness of the suggested strategy is demonstrated by researchers through numerous implementations on real-world high-band picture datasets. The challenges of hyperspectral imaging (HSI) classification, driven by high dimensionality and complex spectral-spatial relationships, demand innovative solutions. Current methodologies, including CNNs and transformer-based networks, suffer from resource demands and interpretability issues, necessitating exploration of combined approaches for enhanced accuracy. In high-band image evaluation and classification applications, the approach delivers state-of-the-art performance and python-implemented model has a 97.8% accuracy rate exceeding previous methods.

Author 1: Manju Sundararajan

Author 2: S. J Grace Shoba

Author 3: Y. Rajesh Babu

Author 4: P N S Lakshmi

Keywords: Deep learning networks; Convolutional Neural Network (CNN); spectral-spatial transformer; adaptive motion optimization; high-band image analysis

Paper 72: Lightweight Cryptographic Algorithms for Medical IoT Devices using Combined Transformation and Expansion (CTE) and Dynamic Chaotic System

Abstract: IoT is growing in prominence as a result of its various applications across many industries. They gather information from the real world and send it over networks. The number of small computing devices, such as RFID tags, wireless sensors, embedded devices, and IoT devices, has increased significantly in the last few years. They are anticipated to produce enormous amounts of sensitive data for the purpose of controlling and monitoring. The security of those devices is crucial because they handle precious private data. An encryption algorithm is required to safeguard these delicate devices. The performance of devices is hampered by traditional encryption ciphers like RSA or AES, which are costly and easy to crack. In the realm of IoT security, lightweight image encryption is crucial. For image encryption, the majority of currently used lightweight techniques use separate pixel values and position modifications. These kinds of schemes are limited by their high vulnerability to cracking. This paper introduces a Lightweight cryptography (LWC) algorithm for medical IoT devices using Combined Transformation and Expansion (CTE) and Dynamic Chaos System. The suggested system is evaluated in terms of cross-entropy, UACI, and NPCR. As demonstrated by the experimental results, the suggested system is ideal for medical IoT systems and has very high encryption and decryption efficiency. The proposed system is characterized by its low memory usage and simplicity.

Author 1: Abdul Muhammed Rasheed

Author 2: Retnaswami Mathusoothana Satheesh Kumar

Keywords: Internet of Things (IoT); data transmission; data security; medical IoT devices; lightweight cryptography; encryption; decryption

Paper 73: A Novel Graph Convolutional Neural Networks (GCNNs)-based Framework to Enhance the Detection of COVID-19 from X-Ray and CT Scan Images

Abstract: The constant need for robust and efficient COVID-19 detection methodologies has prompted the exploration of advanced techniques in medical imaging analysis. This paper presents a novel framework that leverages Graph Convolutional Neural Networks (GCNNs) to enhance the detection of COVID-19 from CT scan and X-Ray images. Hence, the GCNN parameters were tuned by the hybrid optimization to gain a more exact detection. Therefore, the novel technique known as Hybrid NADAM Graph Neural Prediction (NAGNP). The framework is designed to achieve efficiency through a hybrid optimization strategy. The methodology involves constructing graph representations from Chest X-ray or CT scan images, where nodes encapsulate critical image patches or regions of interest. These graphs are fed into GCNN architectures tailored for graph-based data, facilitating intricate feature extraction and information aggregation. A hybrid optimization approach is employed to optimize the model's performance, encompassing fine-tuning of GCNN hyperparameters and strategic model optimization techniques. Through rigorous evaluation and validation using diverse datasets, our framework demonstrates promising results in accurate and efficient COVID-19 diagnosis. Integrating GCNNs and hybrid optimization presents a viable pathway toward reliable and practical diagnostic tools in combating the ongoing pandemic.

Author 1: D. Raghu

Author 2: Hrudaya Kumar Tripathy

Author 3: Raiza Borreo

Keywords: COVID-19 detection; Graph Neural Networks; X-ray; CT scan images; hybrid optimization; medical imaging analysis; diagnostic tools; pandemic response

Paper 74: A Smart AI Framework for Backlog Refinement and UML Diagram Generation

Abstract: In Agile development, it is crucial to refine the backlog to prioritize tasks, resolve problems quickly, and align development efforts with project goals. Automated tools can help in this process by generating Unified Modeling Language (UML) diagrams, allowing teams to work more efficiently with a clear understanding and communicate product requirements. This paper presents an automated approach to Agile methodology which refines backlogs by detecting duplicate user stories and clustering them. Following the refinement process, our approach generates UML diagrams automatically for each cluster, including both class and use case diagrams. Our method is based on machine learning and natural language processing techniques. To implement our approach, we developed a tool that selects the user stories file, groups them by actor, and employs the unsupervised k-means algorithm to form clusters. After that, we used Sentence Bidirectional Encoder Representations from Transformers (SBERT) to measure the similarity between user stories in a cluster. The tool highlights the most similar user stories and facilitates the decision to delete or keep them. Additionally, our approach detects similar or duplicate use cases in the UML use case diagram, making it more convenient for computer system designers. We evaluated our approach on a set of case studies using different performance measures. The results demonstrated its effectiveness in detecting duplicate user stories in the backlog and duplicate use cases. Our automated approach not only saves time and reduces errors, but it also improves collaboration between team members. With an automatic generation of UML diagrams from user stories, all team members can understand product requirements clearly and consistently, regardless of their technical expertise.

Author 1: Samia NASIRI

Author 2: Mohammed LAHMER

Keywords: Artificial intelligence; NLP; Agile methodology; UML

Paper 75: Challenges and Solutions of Agile Software Development Implementation: A Case Study Indonesian Healthcare Organization

Abstract: One healthcare organization in Indonesia has implemented Agile software development (ASD) to complete software development. The organization's problems are post-deployment system bugs, and some software development projects must carry over to the following year. This study aims to assess and provide recommendations for improving agile software development by identifying the challenges faced. Research conducts literature reviews on previous research to identify challenges in ASD in several organizations. Research is also conducted using quantitative methods by surveying software development teams to validate implementation challenges and provide recommendations for these challenges. The results of this study were in the form of a survey attended by thirty-one respondents. The study results found that 14 challenges were faced in other organizations, and 11 were faced by one healthcare organization in Indonesia. Healthcare organizations in Indonesia can apply recommendations to make awareness related to understanding agile software development culture and make adjustments to project documentation by aligning with agile values.

Author 1: Ulfah Nur Mukharomah

Author 2: Teguh Raharjo

Author 3: Ni Wayan Trisnawaty

Keywords: Agile Software Development; challenge solutions; IT projects; information technology; application implementation; healthcare organization; Literature Review

Paper 76: The Bi-Level Particle Swarm Optimization for Joint Pricing in a Supply Chain

Abstract: This study examines the integration of pricing and lot-sizing strategies within a system comprising only one producer and retailer. The adoption of a bi-level programming technique is justified in establishing a bi-level joint pricing model guided by the producer owing to the hierarchical nature of the supply chain. This problem maximizes manufacturer and retailer profitability by setting the wholesale quantity, lot size, and retail price simultaneously. We created a bi-level particle swarm optimization to solve bi-level programming challenges. This algorithm effectively addresses BLPPS by eliminating the need for any priori assumptions about the conditions of the problem. The bi-level particle swarm optimization algorithm demonstrated a commendable level of efficacy when applied to a set of eight benchmark bi-level issues. The proposed bi-level model was solved using the BPSO and analyzed using experimental data.

Author 1: Umar Mansyuri

Author 2: Andreas Tri Panudju

Author 3: Helena Sitorus

Author 4: Widya Spalanzani

Author 5: Nunung Nurhasanah

Author 6: Dedy Khaerudin

Keywords: Bi-Level algorithm; joint pricing; optimization; particle swarm optimization; supply chain

Paper 77: Deep Learning Approach for Workload Prediction and Balancing in Cloud Computing

Abstract: Cloud Computing voted as one of the most revolutionized technologies serving huge user demand engrosses a prominent place in research. Despite several parameters that influence the cloud performance, factors like Workload prediction and scheduling are triggering challenges for researchers in leveraging the system proficiency. Contributions by practitioners given workload prophesy left scope for further enhancement in terms of makespan, migration efficiency, and cost. Anticipating the future workload in due to avoid unfair allocation of cloud resources is a crucial aspect of efficient resource allocation. Our work aims to address this gap and improve efficiency by proposing a Deep Max-out prediction model, which predicts the future workload and facilitates workload balancing paving the path for enhanced scheduling with a hybrid Tasmanian Devil-assisted Bald Eagle Search (TES) optimization algorithm. The results evaluated proved that the TES scored efficiency in makespan with 16.342%, and migration efficiency of 14.75% over existing approaches like WACO, MPSO, and DBOA (Weighted Ant Colony Optimization Modified Particle Swarm Optimization, Discrete Butterfly Optimization Algorithm). Similarly, the error analysis during the evaluation of prediction performance has been figured out using different approaches like MSE, RMSE, MAE, and MSLE, among which our proposed method overwhelms with less error than the traditional methods.

Author 1: Syed Karimunnisa

Author 2: Yellamma Pachipala

Keywords: Task scheduling; virtual machines; optimization; workload prediction; migration; QoS

Paper 78: Estimating Coconut Yield Production using Hyperparameter Tuning of Long Short-Term Memory

Abstract: Coconut production is one of the significant and main sources of revenue in India. In this research, an Auto-Regressive Integrated Moving Average (ARIMA)-Improved Sine Cosine Algorithm (ISCA) with Long Short-Term Memory (LSTM) is proposed for coconut yield production using time series data. It is used for converting non-stationary data to stationary time series data by applying differences. The Holt-Winter Seasonal Method is the Exponential Smoothing variations utilized for seasonal data. The time-series data are given as the input to the LSTM classifier to classify the yield production and the LSTM model is tuned by hyperparameter using Improved Sine Cosine Algorithm (ISCA). In basic SCA, parameter setting and search precision are crucial and the modified SCA improves the coverage speed and search precision of the algorithm. The model’s performance is estimated by utilizing R2, Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Square Error (RMSE) with date on yield from 2011-2021 by categorizing yearly production into 120 records and eight million nuts. The outcomes display that the LSTM-ISCA offers values of 0.38, 0.126, 0.049 and 0.221 for R2, MAE, MSE and RMSE metrics, which offer a precise yield production when related to other models.

Author 1: Niranjan Shadaksharappa Jayanna

Author 2: Raviprakash Madenur Lingaraju

Keywords: Auto-regressive integrated moving average; coconut yield production; improved sine cosine algorithm; long short-term memory; time series

Paper 79: Integrating Lesk Algorithm with Cosine Semantic Similarity to Resolve Polysemy for Setswana Language

Abstract: Word Sense Disambiguation (WSD) serves as an intermediate task for enhancing text understanding in Natural Language Processing (NLP) applications, including machine translation, information retrieval, and text summarization. Its role is to enhance the effectiveness and efficiency of these applications by ensuring the accurate selection of the appropriate sense for polysemous words in diverse contexts. This task is recognized as an AI-complete problem, indicating its longstanding complexity since the 1950s. One of the earliest proposed solutions to address polysemy in NLP is the Lesk algorithm, which has seen various adaptations by researchers for different languages over the years. This study proposes a simplified, Lesk-based algorithm to resolve polysemy for Setswana. Instead of combinatorial comparisons among candidate senses that Lesk is based on that cause computational complexity, this study models word sense glosses using Bidirectional Encoder Representations from Transformers (BERT) and Cosine similarity measure, which have been proven to achieve optimal performance in WSD. The proposed algorithm was evaluated on Setswana and obtained an accuracy of 86.66 and an error rate of 14.34, surpassing the accuracy of other Lesk-based algorithms for other languages.

Author 1: Tebatso Gorgina Moape

Author 2: Oludayo O. Olugbara

Author 3: Sunday O. Ojo

Keywords: Word sense disambiguation; Lesk algorithm; cosine similarity; Bidirectional Encoder Representations from Transformers (BERT)

Paper 80: Design of Emotion Analysis Model IABC-Deep Learning-based for Vocal Performance

Abstract: With the development of deep learning technology, and due to its potential in solving optimization problems with deep structures, deep learning technology is gradually being applied to sentiment analysis models. However, most existing deep learning-based sentiment analysis models have low accuracy issues. Therefore, this study focuses on the issue of emotional analysis in vocal performance. Firstly, based on vocal performance experts and user experience, classify the emotions expressed in vocal performance works to identify the emotional representations of music. On this basis, in order to improve the accuracy of emotion analysis models for deep learning based vocal performance, an improved artificial bee colony algorithm (IABC) was developed to optimize deep neural networks (DNN). Finally, the effectiveness of the proposed deep neural network based on improved artificial bee colony (IABC-DNN) was verified through a training set consisting of 150 vocal performance works and a testing set consisting of 30 vocal performance works. The results indicate that the accuracy of the sentiment analysis model for vocal performance based on IABC-DNN can reach 98%.

Author 1: Zhenjie Zhu

Author 2: Xiaojie Lv

Keywords: Vocal performance; deep learning; Artificial Bee Colony (ABC); emotion analysis model; Deep Neural Network (DNN)

Paper 81: Enhancing the Diagnosis of Depression and Anxiety Through Explainable Machine Learning Methods

Abstract: Diagnosing depression and anxiety involves various methods, including referenda-based approaches that may lack accuracy. However, machine learning has emerged as a promising approach to address these limitations and improve diagnostic accuracy. In this scientific paper, we present a study that utilizes a digital dataset to apply machine learning techniques for diagnosing psychological disorders. The study employs numerical, striatum, and mathematical analytic methodologies to extract dataset features. The Recursive Feature Elimination (RFE) algorithm is used for feature selection, and several classification algorithms, including SVM, decision tree, random forest, logistic regression, and XGBoost, are evaluated to identify the most effective technique for the proposed methodology. The dataset consists of 783 samples from patients with depression and anxiety, which are used to test the proposed strategies. The classification results are evaluated using performance metrics such as accuracy (AC), precision (PR), recall (RE), and F1-score (F1). The objective of this study is to identify the best algorithm based on these metrics, aiming to achieve optimal classification of depression and anxiety disorders. The results obtained will be further enhanced by modifying the dataset and exploring additional machine learning algorithms. This research significantly contributes to the field of mental health diagnosis by leveraging machine learning techniques to enhance the accuracy and effectiveness of diagnosing depression and anxiety disorders.

Author 1: Mai Marey

Author 2: Dina Salem

Author 3: Nora El Rashidy

Author 4: Hazem ELBakry

Keywords: Mental health; Recursive Feature Elimination (RFE); machine learning; Xgboost

Paper 82: An Integrated Arnold and Bessel Function-based Image Encryption on Blockchain

Abstract: Images store large amount of information that are used in visual representation, analysis, and expression of data. Storage and retrieval of images possess a greater challenge to researchers globally. This research paper presents an integrated approach for image encryption and decryption using an Arnold map and first-order Bessel function-based chaos. Traditional methods of image encryption are generally based on single algorithms or techniques, making them vulnerable to various security threats. To address these challenges, our novel method combines the robustness of Arnold transformation with the unique properties of Bessel functions-based chaos. Furthermore, we implemented the decentralized nature of blockchain technology for storing and managing encryption keys securely. By utilizing blockchain's tamper-resistant and transparent ledger, we enhance the integrity and traceability of the encryption process, mitigating the risk of unauthorized access or tampering. The proposed method leverages the chaotic behavior of Bessel function for enhancing security of encryption process. A chaos obtained from first order Bessel function has been utilized for encryption key for encryption after Arnold transformation. The obtained cypher text is stored in blockchain in form of encrypted blocks for secured storage and added security. Experimental evaluations demonstrate the efficiency, effectiveness and robustness of our proposed encryption method when compared with performance of previously developed techniques highlighting the superiority of the proposed method in protecting image data against unauthorized access.

Author 1: Abhay Kumar Yadav

Author 2: Virendra P. Vishwakarma

Keywords: Arnold transformation; block encryption; Bessel functions; blockchain

Paper 83: COOT-Optimized Real-Time Drowsiness Detection using GRU and Enhanced Deep Belief Networks for Advanced Driver Safety

Abstract: Drowsiness among drivers is a major hazard to road safety, resulting in innumerable incidents globally. Despite substantial study, existing approaches for detecting drowsiness in real time continue to confront obstacles, such as low accuracy and efficiency. In these circumstances, this study tackles the critical problems of identifying drowsiness and driver safety by suggesting a novel approach that leverages the combined effectiveness of Gated Recurrent Units (GRU) and Enhanced Deep Belief Networks (EDBN), which is optimised using COOT, a new bird collective-behavioral-based optimisation algorithm. The study begins by emphasising the relevance of sleepiness detection in improving driver safety and the limitations of prior studies in reaching high accuracy in real-time detection. The suggested method tries to close this gap by combining the GRU and EDBN simulations, which are known for their temporal modelling and feature learning capabilities, respectively, to give a comprehensive solution for sleepiness detection. Following thorough experimentation, the suggested technique achieves an outstanding accuracy of around 99%, indicating its efficiency in detecting sleepiness states in real-time driving scenarios. The relevance of this research stems from its potential to greatly reduce the number of accidents caused by drowsy driving, hence improving overall road safety. Furthermore, the use of COOT to optimize the parameters of the GRU and EDBN models adds a new dimension to the research, demonstrating the effectiveness of nature-inspired optimization methodologies for improving the performance of machine learning algorithms for critical applications such as driver safety.

Author 1: Gunnam Rama Devi

Author 2: Hayder Musaad Al-Tmimi

Author 3: Ghadir Kamil Ghadir

Author 4: Shweta Sharma

Author 5: Eswar Patnala

Author 6: B Kiran Bala

Author 7: Yousef A.Baker El-Ebiary

Keywords: Drowsiness detection; driver safety; real-time monitoring; gated recurrent units; enhanced deep belief networks; COOT optimization

Paper 84: A Hybrid Approach with Xception and NasNet for Early Breast Cancer Detection

Abstract: Breast cancer is the most common cancer in women, accounting for 12.5% of global cancer cases in 2020, and the leading cause of cancer deaths in women worldwide. Early detection is therefore crucial to reducing deaths, and recent studies suggest that deep learning techniques can detect breast cancer more accurately than experienced doctors. Experienced doctors can detect breast cancer with only 79% accuracy, while machine learning techniques can achieve up to 91% accuracy (and sometimes up to 97%). To improve breast cancer classification, we conducted a study using two deep learning models, Xception and NasNet, which we combined to achieve better results in distinguishing between malignant and benign tumours in digital databases and cell images obtained from mammograms. Our hybrid model showed good classification results, with an accuracy of over 96.2% and an AUC of 0.993 (99.3%) for mammography data. Remarkably, these results outperformed all other models we compared them with, Top of Form ResNet101 and VGG, which only achieved accuracies of 87%, 88% and 84.4% respectively. Our results were also the best in the field, surpassing the accuracy of other recent hybrid models such as MOD-RES + NasMobile with 89.50% accuracy and VGG 16 + LR with 92.60% accuracy. By achieving this high accuracy rate, our work can make a significant contribution to reducing breast cancer deaths worldwide by helping doctors to detect the disease early and begin treatment immediately.

Author 1: Yassin Benajiba

Author 2: Mohamed Chrayah

Author 3: Yassine Al-Amrani

Keywords: Breast Cancer; CNN; Hybrid Model: Xception; NasNet

Paper 85: Strengthening Sentence Similarity Identification Through OpenAI Embeddings and Deep Learning

Abstract: Discovering similarity between sentences can be beneficial to a variety of systems, including chatbots for customer support, educational platforms, e-commerce customer inquiries, and community forums or question-answering systems. One of the primary issues that online question-answering platforms and customer service chatbots have is the large number of duplicate inquiries that are placed on the platform. In addition to cluttering up the platform, these repetitive queries degrade the content's quality and make it harder for visitors to locate pertinent information. Therefore, it is necessary to automatically detect sentence similarity in order to improve the user experience and quickly match user expectations. The present study makes use of the Quora dataset to construct a framework for similarity discovery in sentence pairs. As part of our research, we have built additional attributes based on textual data for improving the accuracy of similarity prediction. The study investigates several vectorization methods and their influence on accuracy. To convert preprocess text input to a numerical vector, we implemented Word2Vec, FastText, Term Frequency-Inverse Document Frequency (TF-IDF), CountVectorizer (CV), and OpenAI embedding. In order to judge sentence similarity, the embedding offered by several approaches was used with various models, including cosine similarity, Random Forest (RF), AdaBoost, XGBoost, LSTM, and CNN. The result demonstrates that all algorithms trained on OpenAI embedding yield excellent outcomes. The OpenAI-created embedding offers excellent information to models trained on it and has significant potential for capturing sentence similarity.

Author 1: Nilesh B. Korade

Author 2: Mahendra B. Salunke

Author 3: Amol A. Bhosle

Author 4: Prashant B. Kumbharkar

Author 5: Gayatri G. Asalkar

Author 6: Rutuja G. Khedkar

Keywords: OpenAI; embedding; sentence similarity; FastText; Word2Vec; CNN; LSTM; precision; recall; F1-score

Paper 86: Event-based Smart Contracts for Automated Claims Processing and Payouts in Smart Insurance

Abstract: The combination of blockchain technology and smart contracts has become a viable way to expedite claims processing and payouts in the quickly changing insurance industry. Enhancing efficiency, transparency, and reliability for the industry may be achieved by automating certain procedures and initiating them on predetermined triggers, smart contracts that is event-based. Conventional insurance procedures can be laborious, slow, and prone to human mistake, which can cause inefficiencies and delays in the resolution of claims. This research proposes a simplified system that automates the whole claims process from submission to reimbursement by utilizing blockchain technology and smart contracts. The suggested method does away with the requirement for human claim filing by having policyholders' claims automatically triggered by predetermined occurrences. These occurrences might be anything from medical emergencies to natural calamities, enabling prompt and precise claim start. The whole claims process is managed by smart contracts that are programmed with precise triggers and conditions, guaranteeing transaction immutability, security, and transparency. Moreover, reimbursements are carried out automatically after the triggering event has been verified, disregarding conventional bureaucratic processes and drastically cutting down on processing times. This strategy decreases the possibility of fraud and disagreement while also improving operational efficiency by combining self-executing contracts with decentralized ledger technology. Insurance companies and policyholders will both eventually profit from an accelerated, transparent, and reliable claims processing procedure thanks to the use of event-based smart contracts. A Python-implemented system achieving 97.6% accuracy using the proposed method, demonstrates its efficacy and reliability for the given task.

Author 1: Araddhana Arvind Deshmukh

Author 2: Prabhakar Kandukuri

Author 3: Janga Vijaykumar

Author 4: Anna Shalini

Author 5: S. Farhad

Author 6: Elangovan Muniyandy

Author 7: Yousef A.Baker El-Ebiary

Keywords: Blockchain technology; smart contracts; event-based triggers; automated claims processing; transparency and trustworthiness

Paper 87: Real-time Air Quality Monitoring in Smart Cities using IoT-enabled Advanced Optical Sensors

Abstract: Air quality control has drawn a lot of attention from both theoretical research and practical application due to the air pollution problem's increasing severity. As urbanization accelerates, the need for effective air quality monitoring in smart cities becomes increasingly critical. Traditional methods of air quality monitoring often involve stationary monitoring stations, providing limited coverage and outdated data. This study proposes an Internet of Things (IoT) centred framework equipped with inexpensive devices to monitor pollutants vital to human health, in line with World Health Organization recommendations, in response to the pressing issue of air pollution and its increased importance. The hardware development entails building a device that can track significant contamination percentages. Ammonia, carbon monoxide, nitrogen dioxide, PM2.5 and PM10 particulate matter, ozone, and nitrogen dioxide. The gadget is driven by the ESP-WROOM-32 microcontroller, which has Bluetooth and Wi-Fi capabilities for easy data connection to a cloud server. It uses PMSA003, MICS-6814, and MQ-131 sensors. The gadget activates indicators when a pollutant concentration exceeds the allowable limit, enhancing its software to enable immediate response and intervention. This work leverages the robust cloud architecture of Amazon Web Server (AWS) to integrate it into the system and improve accessibility and data control. This combination no longer just ensures data preservation but also enables real-time tracking and analysis, which adds to a comprehensive and preventive strategy for reducing air pollution and preserving public health. With an RMSE score of 3.7656, the Real-Time Alerts with AWS Integration model—which was built in Python—has the lowest value.

Author 1: Anushree A. Aserkar

Author 2: Sanjiv Rao Godla

Author 3: Yousef A.Baker El-Ebiary

Author 4: Krishnamoorthy

Author 5: Janjhyam Venkata Naga Ramesh

Keywords: Internet of Things (IoT); air quality control; low-cost sensors; ESP-WROOM-32 microcontroller; Amazon Web Server (AWS)

Paper 88: DeepCardioNet: Efficient Left Ventricular Epicardium and Endocardium Segmentation using Computer Vision

Abstract: In the realm of medical image analysis, accurate segmentation of cardiac structures is essential for accurate diagnosis and therapy planning. Using the efficient Attention Swin U-Net architecture, this study provides DEEPCARDIONET, a novel computer vision approach for effectively segmenting the left ventricular epicardium and endocardium. The paper presents DEEPCARDIONET, a cutting-edge computer vision method designed to efficiently separate the left ventricular epicardium and endocardium in medical pictures. The main innovation of DEEPCARDIONET is that it makes use of the Attention Swin U-Net architecture, a state-of-the-art framework that is well-known for its capacity to collect contextual information and complicated attributes. Specially designed for the segmentation task, the Attention Swin U-Net guarantees superior performance in identifying the relevant left ventricular characteristics. The model's ability to identify positive instances with high precision and a low false positive rate is demonstrated by its good sensitivity, specificity, and accuracy. The Dice Similarity Coefficient (DSC) illustrates the improved performance of the proposed method in addition to accuracy, showing how effectively it captures spatial overlaps between predicted and ground truth segmentations. The model's generalizability and performance in a variety of medical imaging contexts are demonstrated by its application and evaluation across many datasets. DEEPCARDIONET is an intriguing method for enhancing cardiac picture segmentation, with potential applications in clinical diagnosis and treatment planning. The proposed method achieves an amazing accuracy of 99.21% by using a deep neural network architecture, which significantly beats existing models like TransUNet, MedT, and FAT-Net. The implementation, which uses Python, demonstrates how versatile and useful the language is for the scientific computing community.

Author 1: Bukka Shobharani

Author 2: S Girinath

Author 3: K. Suresh Babu

Author 4: J. Chenni Kumaran

Author 5: Yousef A.Baker El-Ebiary

Author 6: S. Farhad

Keywords: DeepCardioNet; attention swin U-Net; ventricular epicardium; endocardium; computer vision approach

Paper 89: Enhancing HCI Through Real-Time Gesture Recognition with Federated CNNs: Improving Performance and Responsiveness

Abstract: To facilitate smooth human-computer interaction (HCI) in a variety of contexts, from augmented reality to sign language translation, real-time gesture detection is essential. In this paper, researchers leverage federated convolutional neural networks (CNNs) to present a novel strategy that tackles these issues. By utilizing federated learning, authors may cooperatively train a global CNN model on several decentralized devices without sharing raw data, protecting user privacy. Using this concept, researchers create a federated CNN architecture designed for real-time applications including gesture recognition. This federated approach enables continuous model refinement and adaption to various user behaviours and environmental situations by pooling local model updates from edge devices. This paper suggests improvements to the federated learning system to maximize responsiveness and speed. To lessen the probability of privacy violations when aggregating models, this research uses techniques like differential privacy. Additionally, to reduce communication overhead and quicken convergence, To incorporate adaptive learning rate scheduling and model compression techniques research show how federated CNN approach may achieve state-of-the-art performance in real-time gesture detection tasks through comprehensive tests on benchmark datasets. In addition to performing better than centralized learning techniques. This approach guarantees improved responsiveness and adaptability to dynamic contexts. Furthermore, federated learning's decentralized architecture protects user confidentiality and data security, which qualifies it for usage in delicate HCI applications. All things considered, the design to propose a viable path forward for real-time gesture detection system advancement, facilitating more organic and intuitive computer-human interactions while preserving user privacy and data integrity. The proposed federated CNN approach achieves a prediction accuracy in real-time gesture detection tasks, outperforming centralized learning techniques while preserving user privacy and data integrity. The proposed framework that achieves prediction accuracy of 98.70% was implemented in python.

Author 1: R. Stella Maragatham

Author 2: Yousef A. Baker El-Ebiary

Author 3: Srilakshmi V

Author 4: K. Sridharan

Author 5: Vuda Sreenivasa Rao

Author 6: Sanjiv Rao Godla

Keywords: Real-time gesture detection; federated convolutional neural networks; privacy-preserving machine learning; adaptive learning rate scheduling; Decentralized human-computer interaction

Paper 90: Advancing Automated and Adaptive Educational Resources Through Semantic Analysis with BERT and GRU in English Language Learning

Abstract: Semantics describe how language and its constituent parts are understood or interpreted. Semantic analysis is the computer analysis of language to derive connections, meaning, and context from words and sentences. In English language learning, dynamic content generation entails developing instructional materials that adjust to the specific requirements of each student, delivering individualized and contextually appropriate information to boost understanding and engagement. To tailor instructional materials to the various requirements of students, dynamic content creation is essential in English language learning (ELL). This work is a unique method for automatic and adaptive content production in ELL that uses Gated Recurrent Unit (GRU) and Bidirectional Encoder Representations from Transformers (BERT) together. The suggested approach uses BERT enabling content selection, adaption, and adaptive educational content production, and GRU for semantic extraction of features and contextual information captured from textual input. The article presents a novel approach to creating automated and adaptable educational tools for ELL that uses GRU for semantic feature extraction. Using persuasive essays collected in the PERSUADE 2.0 corpus annotated with discourse components and competency scores, this is an extensive dataset. After extensive testing, this approach shows outstanding outcomes, with high accuracy reaching 97% when compared to the current Spiking Neural Network (SNN) &Convolutional Neural Network (CNN), Logistic Regression (LR), and Convolutional Bidirectional Recurrent Neural Network (CBRNN). Python is used to implement the suggested work. The suggested strategy improves ELL engagement and understanding by providing individualized, contextually appropriate learning resources to each student. In addition, the flexibility of the system allows for real-time modifications to suit the changing needs and preferences of the learners. By providing instructors and students in a variety of educational contexts with a scalable and effective approach, this study advances automated content development in ELL. The model architecture will be improved in the future, along with the application's expansion into other domains outside of ELL and the investigation of new language aspects.

Author 1: V Moses Jayakumar

Author 2: R. Rajakumari

Author 3: Sana Sarwar

Author 4: Darakhshan Mazhar Syed

Author 5: Prema S

Author 6: Santhosh Boddupalli

Author 7: Yousef A.Baker El-Ebiary

Keywords: BERT; Content Generation; English Language Learning; Gated Recurrent Unit; Semantic Analysis

Paper 91: Network Security Situation Prediction Technology Based on Fusion of Knowledge Graph

Abstract: It is difficult to accurately reflect different network attack events in real time, which leads to poor performance in predicting network security situations. A knowledge graph-based entity recognition model and entity relationship extraction model was developed for enhancing the reliability and processing efficiency of secure data. Then a knowledge graph-based situational assessment method was introduced, and a network security situational prediction model based on self-attention mechanism and gated recurrent unit was constructed. The study's results showed that the constructed prediction model achieved stable mean square error values of approximately 0.0127 and 0.0136 after being trained on the NSL-KDD and CICIDS2017 datasets for 678 and 589 iterations, respectively. The mean square error value was lower due to fewer training iterations compared to other prediction models. The model was embedded into the information security system of an actual Internet company, and the detection accuracy of the number of network attacks was more than 95%. The results of our study indicate that the method used in the study can accurately predict the network security situation and provide technical support for predicting network information security of the same type.

Author 1: Wei Luo

Keywords: Knowledge graph; network security situation; gated recurrent unit; Bayesian attack graph; relationship extraction; relationship recognition

Paper 92: Hybrid Approach for Enhanced Depression Detection using Learning Techniques

Abstract: According to the World Health Organization (WHO), depression affects over 350 million people worldwide, making it the most common health problem. Depression has numerous causes, including fluctuations in business, social life, the economy, and personal relationships. Depression is one of the leading contributors to mental illness in people, which also has an impact on a person's thoughts, behavior, emotions, and general wellbeing. This study aids in the clinical understanding of patients' mental health with depression. The primary objective of research is to examine learning strategies to enhance the effectiveness of depression detection. The proposed work includes ‘Extended- Distress Analysis Interview corpus’ (E-DAIC) label dataset description and proposed methodology. The membership function applies to the Patients Health Questionnaire (PHQ8_Score) for Mamdani Fuzzy depression detection levels, in addition to the study of the hybrid approach. It also reviews the proposed techniques used for depression detection to improve the performance of the system. Finally, we developed the Ensemble- LSRG (Logistic classifier, Support Vector classifier, Random Forest Classifier, Gradient boosting classifier) model, which gives 98.21% accuracy, precision of 99%, recall of 99%, F1 score of 99%, mean squared error of 1.78%, mean absolute error of 1.78%, and R2 of 94.23.

Author 1: Ganesh D. Jadhav

Author 2: Sachin D. Babar

Author 3: Parikshit N. Mahalle

Keywords: Depression detection; machine learning; extended- distress analysis interview corpus; ensemble-LSRG model; mamdani fuzzy

Paper 93: Sustainable Artificial Intelligence: Assessing Performance in Detecting Fake Images

Abstract: Detecting fake images is crucial because they may confuse and influence people into making bad judgments or adopting incorrect stances that might have disastrous consequences. In this study, we investigate not only the effectiveness of artificial intelligence, specifically deep learning and deep neural networks, for fake image detection but also the sustainability of these methods. The primary objective of this investigation was to determine the efficacy and sustainable application of deep learning algorithms in detecting fake images. We measured the amplitude of observable phenomena using effect sizes and random effects. Our meta-analysis of 32 relevant studies revealed a compelling effect size of 1.7337, indicating that the model's performance is robust. Despite this, some moderate heterogeneity was observed (Q-value = 65.5867; I2 = 52.7344%). While deep learning solutions such as CNNs and GANs emerged as leaders in detecting fake images, their efficacy and sustainability were contingent on the nature of the training images and the resources consumed during training and operation. The study highlighted adversarial confrontations, the need for perpetual model revisions due to the ever-changing nature of image manipulations, and data scarcity as technical obstacles. Additionally, the sustainable deployment of these AI technologies in diverse environments was considered crucial.

Author 1: Othman A. Alrusaini

Keywords: Artificial intelligence; image validation; deep learning; deep neural networks; fake images; image forgery; image manipulations

Paper 94: Federated Convolutional Neural Networks for Predictive Analysis of Traumatic Brain Injury: Advancements in Decentralized Health Monitoring

Abstract: Traumatic Brain Injury (TBI) is a significant global health concern, often leading to long-term disabilities and cognitive impairments. Accurate and timely diagnosis of TBI is crucial for effective treatment and management. In this paper, we propose a novel federated convolutional neural network (FedCNN) framework for predictive analysis of TBI in decentralized health monitoring. The framework is implemented in Python, leveraging three diverse datasets: CQ500, RSNA, and CENTER-TBI, each containing annotated brain CT images associated with TBI. The methodology encompasses data preprocessing, feature extraction using gray level co-occurrence matrix (GLCM), feature selection employing the Grasshopper Optimization Algorithm (GOA), and classification using FedCNN. Our approach achieves superior performance compared to existing methods such as DANN, RF and DT, and LSTM, with an accuracy of 99.2%, surpassing other approaches by 1.6%. The FedCNN framework offers decentralized privacy-preserving training across individual networks while sharing model parameters with a central server, ensuring data privacy and decentralization in health monitoring. Evaluation metrics including accuracy, precision, recall, and F1-score demonstrate the effectiveness of our approach in accurately classifying normal and abnormal brain CT images associated with TBI. The ROC analysis further validates the discriminative ability of the FedCNN framework, highlighting its potential as an advanced tool for TBI diagnosis. Our study contributes to the field of decentralized health monitoring by providing a reliable and efficient approach for TBI management, offering significant advancements in patient care and healthcare management. Future research could explore extending the FedCNN framework to incorporate additional modalities and datasets, as well as integrating advanced deep learning architectures and optimization algorithms to further improve performance and scalability in healthcare applications.

Author 1: Tripti Sharma

Author 2: Desidi Narsimha Reddy

Author 3: Chamandeep Kaur

Author 4: Sanjiv Rao Godla

Author 5: R. Salini

Author 6: Adapa Gopi

Author 7: Yousef A.Baker El-Ebiary

Keywords: Traumatic brain injury; federated learning; convolutional neural network; grasshopper optimization algorithm; health monitoring

Paper 95: Enhancing Threat Detection in Financial Cyber Security Through Auto Encoder-MLP Hybrid Models

Abstract: Cyber-attacks have the potential to cause power outages, malfunctions with military equipment, and breaches of sensitive data. Owing to the substantial financial value of the information it contains, the banking sector is especially vulnerable. The number of digital footprints that banks have increases, increasing the attack surface available to hackers. This paper presents a unique approach to improve financial cyber security threat detection by integrating Auto Encoder-Multilayer Perceptron (AE-MLP) hybrid models. These models use MLP neural networks' discriminative capabilities for detection tasks, while also utilizing auto encoders' strengths in collecting complex patterns and abnormalities in financial data. The NSL-KDD dataset, which is varied and includes transaction records, user activity patterns, and network traffic, was thoroughly analysed. The results show that the AE-MLP hybrid models perform well in spotting possible risks including fraud, data breaches, and unauthorized access attempts. Auto encoders improve the accuracy of threat detection methods by efficiently compressing and rebuilding complicated data representations. This makes it easier to extract latent characteristics that are essential for differentiating between normal and abnormal activity. The approach is implemented with Python software. The recommended Hybrid AE+MLP approach shows better accuracy with 99%, which is 13.16% more sophisticated, when compared to traditional approach. The suggested approach improves financial cyber security systems' capacity for prediction while also providing scalability and efficiency while handling massive amounts of data in real-time settings.

Author 1: Layth Almahadeen

Author 2: Ghayth ALMahadin

Author 3: Kathari Santosh

Author 4: Mohd Aarif

Author 5: Pinak Deb

Author 6: Maganti Syamala

Author 7: B Kiran Bala

Keywords: Financial cyber security; auto encoder; multilayer perceptron; threat detection; hybrid models

Paper 96: Basketball Free Throw Posture Analysis and Hit Probability Prediction System Based on Deep Learning

Abstract: With the continuous progress of basketball technology and tactics, educators need to adopt new teaching methods to cultivate high-quality athletes who meet the needs of modern basketball development. In basketball teaching, the accuracy of free throw techniques directly affects teaching effectiveness. Therefore, the automated prediction of free throw hits is of great significance for reducing manual labor and improving training efficiency. In order to automatically predict the free throw hits and reduce manual fatigue, the study conducts an in-depth analysis for the criticality of free throw in basketball. In this study, the target detection model of target basketball players is constructed based on YOLOv5 and CBAM, and the basketball free throw hit prediction model is constructed based on the OpenPose algorithm. The main quantitative results showed that the proposed model could accurately recognize the athlete posture in free throw actions and save them as video frames in practical applications. Specifically, when using the free throw keyframe limb angle as features, the model achieved a prediction accuracy of 71% and a recall rate of 86% in internal testing. In external testing, the prediction accuracy was improved to 89% and the recall rate was 77%. In addition, combining the relative position difference and angle characteristics of joint points, the accuracy of internal testing was significantly improved to 80%, and the recall rate was increased to 96%. The accuracy of external testing was improved to 95%, with a recall rate of 75%. The experimental results showed that the various functional modules of the system basically meet the expectations, confirming that the basketball penalty posture analysis and hit probability prediction system based on deep learning can effectively assist basketball teaching and meet the practical teaching application needs. The contribution of the research lies in providing a scientific basketball free throw training tool, which helps coaches and athletes better understand and improve free throw techniques, thereby improving free throw hits accuracy. Meanwhile, this study also provides new theoretical and practical references for the application of deep learning in motor skill analysis and training, which has potential value for updating the basketball education system and reducing teacher workload.

Author 1: Yuankai Luo

Author 2: Yan Peng

Author 3: Juan Yang

Keywords: Deep learning; CBAM; OpenPose; Free throws; Posture analysis

Paper 97: Enhancing IoT Network Security: ML and Blockchain for Intrusion Detection

Abstract: Given the proliferation of connected devices and the evolving threat landscape, intrusion detection plays a pivotal role in safeguarding IoT networks. However, traditional methodologies struggle to adapt to the dynamic and diverse settings of IoT environments. To address these challenges, this study proposes an innovative framework that leverages machine learning, specifically Red Fox Optimization (RFO) for feature selection, and Attention-based Bidirectional Long Short-Term Memory (Bi-LSTM). Additionally, the integration of blockchain technology is explored to provide immutable and tamper-proof logs of detected intrusions, bolstering the overall security of the system. Previous research has highlighted the limitations of conventional intrusion detection techniques in IoT networks, particularly in accommodating diverse data sources and rapidly evolving attack strategies. The attention mechanism enables the model to concentrate on pertinent features, enhancing the accuracy and efficiency of anomaly and malicious activity detection in IoT traffic. Furthermore, the utilization of RFO for feature selection aims to reduce data dimensionality and enhance the scalability of the intrusion detection system. Moreover, the inclusion of blockchain technology enhances security by ensuring the integrity and immutability of intrusion detection logs. The proposed framework is implemented using Python for machine learning tasks and Solidity for blockchain development. Experimental findings demonstrate the efficacy of the approach, achieving a detection accuracy of approximately 98.9% on real-world IoT datasets. These results underscore the significance of the research in advancing IoT security practices. By amalgamating machine learning, optimization techniques, and blockchain technology, this framework provides a robust and scalable solution for intrusion detection in IoT networks, fostering improved efficiency and security in interconnected environments.

Author 1: N. Sunanda

Author 2: K. Shailaja

Author 3: Prabhakar Kandukuri

Author 4: Krishnamoorthy

Author 5: Vuda Sreenivasa Rao

Author 6: Sanjiv Rao Godla

Keywords: Intrusion detection; IoT networks; machine learning; random forest; red fox optimization; blockchain technology

Paper 98: Unveiling Spoofing Attempts: A DCGAN-based Approach to Enhance Face Spoof Detection in Biometric Authentication

Abstract: Face spoofing attacks have become more dangerous as biometric identification has become more widely used. Through the utilisation of false facial photographs, attackers seek to fool systems in these assaults, endangering the security of biometric authentication devices and perhaps allowing unauthorized access to private information. Effectively recognizing and thwarting such spoofing attacks is critical to the dependability and credibility of biometric identification systems in a variety of applications. This research seeks to offer a unique strategy that uses Deep Convolutional Generative Adversarial Networks (DCGANs) to improve face spoof detection in order to counter the challenge provided by face spoofing assaults. In order to strengthen the security of biometric authentication systems in applications like identity verification, access control, and mobile device unlocking, the goal is to increase the accuracy and effectiveness of facial spoof detection. The training dataset is then supplemented with these artificial images, which strengthens the face spoof detection system's resilience. More accurate face spoofing is made possible by the strategy that leverages the discriminative characteristics obtained throughout the process to train the discriminator network employing adversarial learning to discriminate between actual and fake images. Experiments on the CelebFacesAttributes (CelebA) datasets show how effective the suggested method is over traditional techniques. The suggested technique outperforms conventional methods and achieves an astounding accuracy of 99.1% in face-spoof detection systems. The system exhibits impressive precision in differentiating between real and fake faces through the efficient use of artificial intelligence and adversarial learning. This effectively decreases the possibility of unwanted access and enhances the overall dependability of biometric authentication methods.

Author 1: Vuda Sreenivasa Rao

Author 2: Shirisha Kasireddy

Author 3: Annapurna Mishra

Author 4: R. Salini

Author 5: Sanjiv Rao Godla

Author 6: Khaled Bedair

Keywords: Biometric authentication systems; deep convolutional generative adversarial networks; face spoof detection; synthetic image generation; unauthorized access

Paper 99: A Novel Proposal for Improving Economic Decision-Making Through Stock Price Index Forecasting

Abstract: The non-stationary, non-linear, and extremely noisy nature of stock price time series data, which are created from economic factors and systematic and unsystematic risks, makes it difficult to make reliable predictions of stock prices in the securities market. Conventional methods may improve forecasting accuracy, but they can additionally complicate the computations involved, increasing the likelihood of prediction errors. To address these issues, a novel hybrid model that combines recurrent neural networks and grey wolf optimization was introduced in the current study. The suggested model outperformed other models in the study with high efficacy, minimal error, and peak performance. Utilizing data from Alphabet stock spanning from June 29, 2023, to January 1, 2015, the effectiveness of the hybrid model was assessed. The gathered information comprised daily prices and trading volume. The outcomes showed that the suggested model is a reliable and effective method for analyzing and forecasting the time series of the financial market. The suggested model is additionally particularly well-suited to the volatile stock market and outperforms other recent strategies in terms of forecasting accuracy.

Author 1: Xu Yao

Author 2: Weikang Zeng

Author 3: Lei Zhu

Author 4: Xiaoxiao Wu

Author 5: Di Li

Keywords: Hybrid model; recurrent neural networks; grey wolf optimization; stock price prediction

Paper 100: Optimization Method for Digital Twin Manufacturing System Based on NSGA-II

Abstract: In the wave of industrial modernization, a concept that comprehensively covers the product lifecycle has been proposed, namely the digital twin manufacturing system. The digital twin manufacturing system can conduct three-dimensional simulation of the workshop, thereby achieving dynamic scheduling and energy efficiency optimization of the workshop. The optimization of digital twin manufacturing systems has become a focus of research. In order to reduce power consumption and production time in manufacturing workshops, the study adopted a non-dominated sorting genetic algorithm to improve its elitist retention strategy for the problem of easily falling into local optima. On the ground of the idea of multi-objective optimization, the optimization was carried out with the production time and power consumption of the manufacturing industry as the objectives. The experiment showcased that the improved algorithm outperforms the multi-objective optimization algorithm on the ground of decomposition and the evolutionary algorithm on the ground of Pareto dominance. Compared to the two comparison algorithms, the production time optimization effect and power consumption optimization effect of different numbers of devices were 11.12%-21.37% and 2.14%-6.89% higher, respectively. The optimization time of the improved algorithm was 713.5 seconds, that was reduced by 173.8 seconds and 179.8 seconds compared to the other two algorithms, respectively. The total power consumption of the improved optimization model was 2883.7kWs, which was reduced by 32.0kW*s and 45.5kW*s compared to the other two algorithms, respectively. This study proposed a new multi-objective optimization algorithm for the current digital twin manufacturing industry. This algorithm effectively reduces production time and power consumption, and has important guiding significance for manufacturing system optimization in actual production environments.

Author 1: Yu Ding

Author 2: Longhua Li

Keywords: Multi-objective optimization; NSGA-II; Digital twin; Production time; Production energy consumption

Paper 101: Keyword Acquisition for Language Composition Based on TextRank Automatic Summarization Approach

Abstract: It is important to extract keywords from text quickly and accurately for composition analysis, but the accuracy of traditional keyword acquisition models is not high. Therefore, in this study, the Best Match 25 algorithm was first used to preprocess the compositions and evaluate the similarity between sentences. Then, TextRank was used to extract the abstract, construct segmentation and named entity model, and finally verify the research content. The results show that in the performance test, the Best Match 25 similarity algorithm has higher accuracy, recall rate and F1 value, the average running time is only 2182ms, and has the largest receiver working characteristic curve area, which is significantly higher than other models, reaching 0.954. The accuracy of TextRank algorithm is above 90%, the average accuracy of 100 text analysis is 94.23%, the average recall rate and F1 value are 96.67% and 95.85%, respectively. In comparison of the application of the four methods, the research model shows obvious advantages, the average keyword coverage rate is 94.54%, the average processing time of 16 texts is 11.29 seconds, and the average 24-hour memory usage is only 15.67%, which is lower than the other three methods. The experimental results confirm the superiority of the model in terms of keyword extraction accuracy. This research not only provides a new technical tool for language composition teaching and evaluation, but also provides a new idea and method for keyword extraction research in the field of natural language processing.

Author 1: Yan Jiang

Author 2: Chunlin Xiang

Author 3: Lingtong Li

Keywords: Language composition; keywords; best match 25; textrank; digests

Paper 102: Investigating Cooling Load Estimation via Hybrid Models Based on the Radial Basis Function

Abstract: To advance energy conservation in cooling systems within buildings, a pivotal technology known as cooling load prediction is essential. Traditional industry computational models typically employ forward or inverse modeling techniques, but these methods often demand extensive computational resources and involve lengthy procedures. However, artificial intelligence (AI) surpasses these approaches, with its models exhibiting the capability to autonomously discern intricate patterns, adapt dynamically, and enhance their performance as data volumes increase. AI models excel in forecasting cooling loads, accounting for various factors like weather conditions, building materials, and occupancy. This results in agile and responsive predictions, ultimately leading to heightened energy efficiency. The dataset of this study, which comprised 768 samples, was derived from previous studies. The primary objective of this study is to introduce a novel framework for the prediction of Cooling Load via integrating the Radial Basis Function (RBF) with 2 innovative optimization algorithms, specifically the Dynamic Arithmetic Optimization Algorithm (DAO) and the Golden Eagle Optimization Algorithm (GEO). The predictive outcomes indicate that the RBDA prediction model outperforms RBF in cooling load predictions, with RMSE=0.792, approximately half as much as those of RBF. Furthermore, the RBDA model's performance, especially in the training phase, confirmed the optimal value of R2=0.993.

Author 1: Sirui Zhang

Author 2: Hao Zheng

Keywords: Cooling load estimation; machine learning; building energy consumption; radial basis functions; dynamic arithmetic optimization algorithm; golden eagle optimization algorithm

Paper 103: Adaptive Target Region Attention Network-based Human Pose Estimation in Smart Classroom

Abstract: In smart classroom environments, problems such as occlusion and overlap make the acquisition of student pose information challenging. To address these problems, a lightweight human pose estimation model with Adaptive Target Region Attention based on Lite-HRNet is proposed for smart classroom scenarios. Firstly, the Deformable Convolutional Encoding Network (DCEN) module is designed to reconstruct the encoding of features through an encoder and then a multi-layer deformable convolutional module is used to adaptively focus on the image region to obtain a feature representation that focuses on the target region of interest of the student subject. Secondly, the Channel And Spatial Attention (CASA) module is designed to attenuate or enhance the feature attention in different regions of the feature map to obtain a more accurate representation of the target feature. Finally, extensive experiments were conducted on the COCO dataset and the smart classroom dataset (SC-Data) to compare the proposed model with the current main popular human pose estimation framework. The experimental results show that the performance of the model reaches 67.5(mAP) in the COCO dataset, which is an improvement of 2.7(mAP) compared to the Lite-HRNet model, and 86.6(mAP) in the SC-Data dataset, which is an improvement of 1.6(mAP) compared to the Lite-HRNet model.

Author 1: Jianwen Mo

Author 2: Guiyun Jiang

Author 3: Hua Yuan

Author 4: Zhaoyu Shou

Author 5: Huibing Zhang

Keywords: Human pose estimation; smart classroom; Lite-HRNet; deformable convolutional encoding network; target region attention

Paper 104: Breast Cancer Classification through Transfer Learning with Vision Transformer, PCA, and Machine Learning Models

Abstract: Breast cancer is a leading cause of death among women worldwide, making early detection crucial for saving lives and preventing the spread of the disease. Deep Learning and Machine Learning techniques, coupled with the availability of diverse breast cancer datasets, have proven to be effective in assisting healthcare practitioners worldwide. Recent advancements in image classification models, such as Vision Transformers and pretrained models, offer promising avenues for breast cancer imaging classification research. In this study, we employ a pretrained Vision Transformer (ViT) model, specifically trained on the ImageNet dataset, as a feature extractor. We combine this with Principal Component Analysis (PCA) for dimensionality reduction and evaluate two classifiers, namely a Multilayer Perceptron (MLP) and a Support Vector Machine (SVM), for breast mammogram image classification. The results demonstrate that the transfer learning approach using ViT, PCA, and an MLP classifier achieves an average accuracy, precision, recall, and F1-score of 98% for the DSMM dataset and 95% for the INbreast dataset, considering the same metrics which are comparable to the current state-of-the-art.

Author 1: Juan Gutierrez-Cardenas

Keywords: Breast cancer; vision transformer; transfer learning; PCA; machine learning

Paper 105: Hybrid Algorithm using Rivest-Shamir-Adleman and Elliptic Curve Cryptography for Secure Email Communication

Abstract: Email serves as the primary communication system in our daily lives, and to bolster its security and efficiency, many email systems employ Public Key Infrastructure (PKI). However, the convenience of email also introduces numerous security vulnerabilities, including unauthorized access, eavesdropping, identity spoofing, interception, and data tampering. This study is primarily focused on examining how two encryption techniques, RSA and ECC, affect the efficiency of secure email systems. Furthermore, the research seeks to introduce a hybrid cryptography algorithm that utilizes both RSA and ECC to ensure security and confidentiality in the context of secure email communication. The research evaluates various performance metrics, including key exchange time, encryption and decryption durations, signature generation, and verification times, to understand how these encryption methods affect the efficiency and efficacy of secure email communication. The experimental findings highlight the advantages of ECC in terms of Key Exchange Time, making it a compelling choice for establishing secure email communication channels. While RSA demonstrates a slight advantage in encryption, decryption, and signature generation for smaller files, ECC's efficiency becomes apparent as file sizes increase, positioning it as a favorable option for handling larger attachments in secure emails. Through the comparison of experiments, it is also concluded that the hybrid encryption algorithm optimizes the key exchange times, encryption efficiency, signature generation and verification times.

Author 1: Kwame Assa-Agyei

Author 2: Kayode Owa

Author 3: Tawfik Al-Hadhrami

Author 4: Funminiyi Olajide

Keywords: RSA; ECC; Advanced Encryption Standard; encryption; decryption; signature generation; verification; key exchange time; hybrid encryption

Paper 106: Federated Machine Learning for Epileptic Seizure Detection using EEG

Abstract: Early seizure detection is difficult with epilepsy. This use of Electroencephalography (EEG) data has proven transformational, however standard centralized machine learning algorithms have privacy and generalization issues. A decentralized approach to epileptic seizure detection using Federated Machine Learning (FML) is presented in this research. The concentration of critical EEG data in conventional models may compromise patient confidentiality. The proposed FML technique trains models using local datasets without sharing raw EEG recordings. Hence the data set used for the model is devoid of noise thus rendering preprocessing unnecessary. Training using decentralized data sources broadens the model's seizure pattern repertoire, improving its adaptability to case heterogeneity. The Federated Machine Learning (FML) model shows that the suggested method for EEG-based epileptic seizure identification is promising for healthcare implementation and deployment. The proposed approach obtains sensitivity, specificity, and accuracy of 98.24%, 99.23%, 99% respectively. The proposed study is validated with the existing literature and the developed model outperforms the existing study.

Author 1: S. Vasanthadev Suryakala

Author 2: T. R. Sree Vidya

Author 3: S. Hari Ramakrishnans

Keywords: Federal Machine Learning (FML); electroencephalography; epileptic seizure; cross-decentralization; health care; sensitivity

Paper 107: Impact of the IoT Integration and Sustainability on Competition Within an Oligopolistic 3PL Market

Abstract: The third party logistics (3PL) sector holds a crucial role in modern supply chains, streamlining the movement of goods and optimizing logistics operations. The 3PL industry’s journey towards digitalization and sustainability reflects a crucial strategy to create an efficient and resilient supply chain. It is increasingly integrating Internet of Things technologies (IoT) within its operations. This latter is a cutting-edge technology widely used in the supply chain realm as it offers numerous advantages namely traceability and real-time decision-making capability. In view of growing concerns for the environment and the social welfare, supply chain actors are seeking to make various initiatives to shift to more sustainable practices. This paper studies the competition within an oligopolistic market of 3PL firms. Through the lens of game theory, we construct a mathematical model where a supply chain composed of n firms competes through pricing, IoT integration efforts and sustainability efforts. Results show that the IoT integration and sustainability efforts impact the pricing decisions of the firm. Moreover, this study highlights how the rivals’ decisions on the IoT integration and sustainability efforts impact the firm’s decision-making processes. Furthermore, a comparison of the model decision variables within a duopoly and an oligopolistic setting is conducted. This paper concludes to the significant impact of the rivals’ strategies on the firm’s decisions and profitability.

Author 1: Kenza Izikki

Author 2: Aziz Ait Bassou

Author 3: Mustapha Hlyal

Author 4: Jamila El Alami

Keywords: Third party logistics; internet of things; sustainability; oligopoly; game theory

Paper 108: Unified Approach for Scalable Task-Oriented Dialogue System

Abstract: Task-oriented dialogue (TOD) systems are currently the subject of extensive research owing to their immense significance in the fields of human-computer interaction and natural language processing. These systems assist users to accomplish certain tasks efficiently. However, most commercial TOD systems rely on handcrafted rules and offer functionalities in a single domain. These systems perform well but are not scalable to adapt multiple domains without manual efforts. Pretrained language models (PLMs) have been popularly applied to enhance these systems via fine-tuning. Recently, large language models (LLMs) have made significant advancements in this field but lack the ability to converse proactively in multiple turns, which is an essential parameter for designing TOD systems. To address these challenges, this paper initially studies the impact of language understanding on the overall performance of a TOD system in a multi-domain environment. Furthermore, to design an efficient TOD system, we propose a unified approach by leveraging LLM with reinforcement learning (RL) based dialogue policy. The experimental results demonstrate that a unified approach using LLM is more promising for scaling the capabilities of TOD systems with prompt adaptive instructions with more user friendly and human-like response generation.

Author 1: Manisha Thakkar

Author 2: Nitin Pise

Keywords: Task-oriented dialogue system; unified; adaptive multi-domain; large language models; prompts

Paper 109: Day Trading Strategy Based on Transformer Model, Technical Indicators and Multiresolution Analysis

Abstract: Stock prices are very volatile because they are affected by infinite number of factors, such as economical, social, political, and human behavior. This makes finding consistently profitable day trading strategy extremely challenging and that is why an overwhelming majority of stock traders loose money over time. Professional day traders, who are very few in number, have a trading strategy that can exploit this price volatility to consistently earn profit from the market. This study proposes a consistently profitable day trading strategy based on price volatility, transformer model, time2vec, technical indicators, and multiresolution analysis. The proposed trading strategy has eight trading systems, each with a different profit-target based on the risk taken per trade. This study shows that the proposed trading strategy results in consistent profits when the profit-target is 1.5 to 3.5 times the risk taken per trade. If the profit-target is not in that range, then it may result in a loss. The proposed trading strategy was compared with the buy-and-hold strategy and it showed consistent profits with all the stocks whereas the buy-and-hold strategy was inconsistent and resulted in losses in half the stocks. Also three of the consistently profitable trading systems showed significantly higher average profits and expectancy than the buy-and-hold trading strategy.

Author 1: Salahadin A. Mohammed

Keywords: Artificial neural network; saudi stock exchange; machine learning; deep learning; transformer model; stock price prediction; time series analysis; technical analysis; multiresolution analysis

Paper 110: Multi-Granularity Feature Fusion for Enhancing Encrypted Traffic Classification

Abstract: Encrypted traffic classification, a pivotal process in network security and management, involves analyzing and categorizing data traffic that has been encrypted for privacy and security. This task demands the extraction of distinctive and robust feature representations from content-concealed data to ensure accurate and reliable classification. Traditional approaches have focused on utilizing either the payload of encrypted traffic or statistical features for more precise classification. While these methods achieve relative success, their limitation lies in not harnessing multi-grained features, thus impeding further advance-ments in encrypted traffic classification capabilities. To tackle this challenge, ET-CompBERT is presented, an innovative framework specifically designed for the fusion of multi-granularity features in encrypted traffic, encompassing both payload and global temporal attributes. The extensive experiments reveal that our approach significantly enhances classification performance in data-rich scenarios (achieving up to a +4.43% improvement in certain cases over existing methods) and establishes state-of-the-art results on training sets with different sizes. The source codes will be released after paper acceptance.

Author 1: Quan Ding

Author 2: Zhengpeng Zha

Author 3: Yanjun Li

Author 4: Zhenhua Ling

Keywords: Encrypted traffic classification; BERT; multi-granularity fusion

Paper 111: Optimization of PID Controller Parameter using the Geometric Mean Optimizer

Abstract: The PID controller is a crucial element in numerous engineering applications. However, a significant challenge with PID lies in selecting optimal parameter values. Conventional methods need extra tunning and may not yield the best performance. In this study, a recently introduced metaheuristic algorithm, Geometric Mean Optimizer (GMO), is employed to identify the most suitable PID parameter values. In conventional methods, a fixed empirical equations are applied to select parameter values of PID. In GMO, there is a wide search space to select the optimal parameter values of PID based on an objective function. The objective function that the GMO seeks to minimize is the Integral of Absolute Error (IAE). GMO is chosen for its effectiveness in balancing exploration and exploitation of the search space, as well as its robustness and scalability. GMO is tested in the context of optimizing PID parameters for an engineering application: DC motor regulations. The results demonstrated GMO’s superiority over comparable algorithms.

Author 1: Osama Abdellatif

Author 2: Mohamed Issa

Author 3: Ibrahim Ziedan

Keywords: Metaheuristics; PID controller; GMO; DC motor

Paper 112: Blockchain-Driven Decentralization of Electronic Health Records in Saudi Arabia: An Ethereum-Based Framework for Enhanced Security and Patient Control

Abstract: In the rapidly evolving landscape of e-HealthCare in Saudi Arabia, enhancing the security and integrity of Electronic Health Records (EHRs) is imperative. Existing systems encounter challenges stemming from centralized storage, vulner-able data integrity, susceptibility to power failures, and issues of ownership by entities other than the patients themselves. Moreover, the sharing of sensitive patient information among anonymous bodies exacerbates the vulnerability of these records. In response to these challenges, this paper advocates for the trans-formative potential of blockchain technology. Blockchain, with its decentralized and distributed architecture, offers a revolutionary approach to communication among network nodes, eliminating the need for a central authority. This paper proposes a solution that places the patient at the forefront, empowering them as the primary controller of their medical data. The research delves into the current state of e-HealthCare in Saudi Arabia, examines the challenges faced by existing EHR systems, and introduces blockchain technology, particularly Ethereum, as a viable and transformative solution. The paper details the use of Ethereum blockchain to secure and manage medical records, with a Public Key Infrastructure (PKI) applied to safeguard the confidentiality of patient information. The decentralized InterPlanetary File System (IPFS) is employed for the secure and resilient storage of encrypted medical records. Additionally, Smart contracts, integral to the Ethereum blockchain, play a central role in automating and enforcing the rules governing access to medical records. Moreover, a Web 3.0 decentralized application (DApp) is developed to provide a user-friendly interface, empowering patients to seamlessly interact with and control access to their health data. At the end, this paper presents a guiding framework for clinicians, policymakers, and academics, illustrating the trans-formative potential of blockchain and associated technologies in revolutionizing EHR management in Saudi Arabia’s healthcare systems.

Author 1: Atef Masmoudi

Author 2: Maha Saeed

Keywords: Blockchain; Ethereum; smart contract; Web 3.0; decentralized application; electronic health records

Paper 113: Automating Tomato Ripeness Classification and Counting with YOLOv9

Abstract: This article proposes a novel solution to the long-standing issue of ripe (or manual) tomato monitoring and counting, often relying on visual inspection, which is both time-consuming, requires a lot of labor and prone to inaccuracies. By leveraging the power of artificial intelligence (AI) and image analysis techniques, a more efficient and precise method for automating this process is introduced. This approach promises to significantly reduce labor requirements while enhancing accuracy, thus improving overall quality and productivity. In this study, we explore the application of the latest version of YOLO (You Only Look Once), specifically YOLOv9, in automating the classification of tomato ripeness levels and counting tomatoes. To assess the performance of the proposed model, the study employs standard evaluation metrics including Precision, Recall, and mAP50. These metrics provide valuable insights into the model’s ability to accurately detect and count tomatoes in real-world scenarios. The results indicate that the YOLOv9-based model achieves superior performance, as evidenced by the following evaluation metrics: Precision: 0.856, Recall: 0.832, and mAP50: 0.882. By leveraging YOLOv9 and comprehensive evaluation metrics, this research aims to provide a robust solution for automating tomato monitoring processes. Additionally, by offering the future integration of robotics, the collection phase can further optimize efficiency and enable the expansion of cultivation areas.

Author 1: Hoang-Tu Vo

Author 2: Kheo Chau Mui

Author 3: Nhon Nguyen Thien

Author 4: Phuc Pham Tien

Keywords: Tomato monitoring; manual counting; Artificial Intelligence (AI); Image analysis techniques; YOLO; YOLOv9

Paper 114: Harnessing AI to Generate Indian Sign Language from Natural Speech and Text for Digital Inclusion and Accessibility

Abstract: Sign language is the fundamental mode of communication for those who are deaf and mute, as well as for individuals with hearing impairments. Regrettably, there has been a dearth of research on Indian Sign Language, primarily due to the lack of adequate grammar and regional variations in such language. Consequently, research in this area has been limited. The primary objective of our research is to develop a sophisticated speech/ text-to-Indian sign language conversion system that employs advanced 3D modeling techniques to display sign language motions. Our research is motivated by our desire to promote effective communication between hearing and hearing-impaired individuals in India. The proposed model integrates Automatic Speech Recognition (ASR) technology, which effectively transforms spoken words into text, and leverages 3D modeling techniques to generate corresponding sign language motions. We have conducted a comprehensive study of the grammar of Indian Sign Language, which includes identifying sentence structure and signs that represent the tense of the subject. It is noteworthy that the sentence structure of Indian Sign Language follows the Subject-Object-Verb sequence, in contrast to spoken language, which follows the Subject-Verb-Object structure. To enhance user experience as well as digital inclusion and accessibility, the research incorporates user-friendly and simple interfaces that allow individuals to interact effortlessly with the system intuitively. The model/ system is equipped to receive speech input through a microphone/ text and provide immediate feedback through 3D-modeled videos that display the generated sign language gestures and has achieved 99.2% accuracy. Our main goal is to promote digital inclusion and improve accessibility and enhance the user experience.

Author 1: Parul Yadav

Author 2: Puneet Sharma

Author 3: Pooja Khanna

Author 4: Mahima Chawla

Author 5: Rishi Jain

Author 6: Laiba Noor

Keywords: Sign language generation; automatic speech recognition; speech-to-indian sign language; indian sign language; digital inclusion and accessibility

Paper 115: Developing a Patient-Centric Healthcare IoT Platform with Blockchain and Smart Contract Data Management

Abstract: The Internet of Things (IoT) has been rapidly integrated into various industries, with healthcare emerging as a key area of impact. A notable development in this sector is the IoHT-MBA system, a specialized Internet of Healthcare Things (IoHT) framework. This system utilizes a microservice approach combined with a brokerless architecture, efficiently tackling issues like data gathering, managing users and devices, and controlling devices remotely. Despite its effectiveness, there’s a growing need to improve the privacy and control of patient data. To address this, we propose an enhanced version of the IoHT-MBA system, incorporating blockchain technology, specifically through the use of Hyperledger Fabric. This integration aims to create a more secure, transparent, and patient-centric data management platform. The system enables patients to oversee their peripheral devices, such as smartphones and sensors. These devices are integrated as part of the edge layer of the IoHT, contributing to a decentralized storage service. In our model, data is primarily retained on user devices, with only summarized data being communicated to service providers and recorded on the blockchain. This approach significantly boosts data privacy and user control. Access to user data is strictly regulated and must align with the patient’s privacy conditions, which are established through smart contracts, thus providing an additional layer of security and transparency. We have conducted an evaluation of our blockchain-enhanced platform using key theories in microservice and brokerless architecture, such as Round Trip Time and Broken Connection Test Cases. Additionally, we’ve performed tests on data generation and queries using Hyperledger Caliper. The results confirm the strength and efficiency of our blockchain-integrated system in the healthcare IoT domain.

Author 1: Duc B. T

Author 2: Trung P. H. T

Author 3: Trong N. D. P

Author 4: Phuc N. T

Author 5: Khoa T. D

Author 6: Khiem H. G

Author 7: Nam B. T

Author 8: Bang L. K

Keywords: Medical test result; blockchain; smart contract; NFT; Ethereum; Fantom; polygon; binance smart chain

Paper 116: GROCAFAST: Revolutionizing Grocery Shopping for Seamless Convenience and Enhanced User Experience

Abstract: This paper presents the Smart Grocery Shopping system (GROCAFAST), a system for optimizing the grocery shopping experience and improving efficiency for shoppers. The GROCAFAST system consists of a mobile app and a server component. The mobile app allows shoppers to create, manage, and update grocery lists while providing store navigation assistance. The server component processes data, generates optimized route maps, maintains an inventory database, and facilitates the online chat room. Unlike existing grocery shopping systems, GROCAFAST is cost-effective as it does not rely on any extra infrastructure and reduces both shopping time and walking steps. GROCAFAST utilizes Dijkstra’s algorithm to efficiently guide shoppers through the store, minimizing the time needed to visit all aisles containing their desired items. The user-friendly interface and time-saving features make grocery shopping more efficient and enjoyable. The evaluation results demonstrate that GROCAFAST reduces the total shopping time by 67.6% when compared to a traditional approach that mimics the way shoppers visit a grocery store, browse aisles, and select items. It also reduces the walking steps by 59%.

Author 1: Abeer Hakeem

Author 2: Layan Fakhurji

Author 3: Raneem Alshareef

Author 4: Elaf Aloufi

Author 5: Manar Altaiary

Author 6: Afraa Attiah

Author 7: Linda Mohaisen

Keywords: Grocery shopping app; route map; grocery shopping experience; dijkstra’s algorithm

Paper 117: New Trust Management Scheme Based on Blockchain and KNN Reinforcement Learning Algorithm

Abstract: There has been a continual rise in the quantity of smart and autonomous automobiles in recent decades. the effectiveness of communication among vehicles in Vehicular Ad-hoc Networks (VANET) is critical for ensuring the safety of drivers’ lives. the primary objective of VANET is to share critical information regarding life-threatening events, such as traffic jams and accident alerts in a timely and accurate manner. Nevertheless, typical VANETs encounter several security issues involving threats to confidentiality, integrity, and availability. This paper proposes a new decentralized and tamper-resistant scheme for privacy preservation. We designed a new trust management system that utilizes blockchain technology. We strive to establish trust between vehicles and infrastructure and preserve privacy by guaranteeing the authenticity and integrity of the information exchanged in VANETS. Our proposal adopts the principles of reinforcement learning to dynamically evaluate and allocate trust scores to vehicles and infrastructure based on their behavior. The scheme’s performance has been evaluated based on key metrics. The results show that our new system provides an effective behavior management technique while preserving vehicle privacy.

Author 1: Ahdab Hulayyil Aljohani

Author 2: Abdulaziz Al-shammri

Keywords: Vehicular Ad hoc Networks (VANETs); Blockchain; trust management; reinforcement learning algorithm; privacy preservation; network security

Paper 118: An Efficiency Hardware Design for Lane Detector Systems

Abstract: The Hough Transform (HT) algorithm is a popular method for lane detection based on the 'voting' process to extract complete lines. The voting process is derived from the HT algorithm and then executed in parameter space (ρ, θ) to identify the 'votes' with the highest count, meaning that image points with pairs of angle θ and distance ρ corresponding to those 'votes' lie on the same line. However, this algorithm requires significant memory and computational complexity. In this paper, we propose a new algorithm for the Hough Space (HS) by utilizing parameter-ization (Y-intercept, θ) instead of (ρ, θ) parameterization and lane direction. This simplifies the inverse LHT operation and reduces the accumulator's size and computational complexity compared to the standard LHT. We aim to minimize processing time per frame for real-time processing. Our implementation operates at a frequency of 250MHz, and the processing time for each frame with a resolution of 1024x1024 is 4.19ms, achieving an accuracy of 85.49%. This design is synthesized on the Virtex-7 VC707 FPGA.

Author 1: Duc Khai Lam

Keywords: FPGA; Hough transform; look up table; lane detector; autonomous vehicle

Paper 119: Predictor Model for Chronic Kidney Disease using Adaptive Gradient Clipping with Deep Neural Nets

Abstract: This research aims to develop computer vision based predictive model for the three prominent kidney ailments namely Cyst, Stone, and Tumor which are common renal disorders that require timely medical intervention. This classification model is tested and trained using the multi-class CT Kidney Dataset which contains 12,446 images collected from PACS (Picture Archiving and Communication System) from different hospitals in Dhaka, Bangladesh. Initial models are build using plain VGG16, ResNet50, and InceptionV3 deep neural nets. Then after clip value filter of ADAM optimizer is applied which results in marginally improved accuracy and at the last Adaptive Gradient Clipping is applied as a replacement of batch norm process and this produces overall best results. The Adaptive Gradient Clipping based model achieves accuracy of 97.15% in VGG16, 99.5% in ResNet50, and 99.23% in InceptionV3. Overall classification metrics are best for ResNet50 and Inception V3 with Adaptive Gradient Clipping technique.

Author 1: Neeraj Sharma

Author 2: Praveen Lalwani

Keywords: CT Kidney; VGG16; ResNet50; InceptionV3; gradient clipping; image processing; multiclass classification

Paper 120: Development of an Educational Robot for Exploring the Internet of Things

Abstract: Educational robots, when integrated into STEM (Science, Technology, Engineering, and Mathematics) education across a range of age groups, serve to enhance learning experiences by facilitating hands-on activities. These robots are particularly instrumental in the realm of Internet of Things (IoT) education, guiding learners from basic to advanced applications. This paper introduces the IoTXplorBot, an open-source and open-design educational robot, developed to foster the learning of IoT concepts in a cost-effective manner. The robot is equipped with a variety of low-cost sensors and actuators and features an interchangeable microcontroller that is compatible with other development boards from the Arduino Nano family. This compatibility allows for diverse programming languages and varied purposes. The robot’s printed circuit board is designed to be user-friendly, even for those with no engineering skills. The proposed board includes additional pins and a breadboard on the robot’s chassis, enabling the extension of the robot with other hardware components. The use of the Arduino board allows learners to leverage all capabilities from Arduino, such as the Arduino IoT cloud, dashboard, online compiler, and project hub. These resources aid in the development of new projects and in finding solutions to encountered problems. The paper concludes with a discussion on the future development of this robot, underscoring its potential for ongoing adaptation and improvement.

Author 1: Zhumaniyaz Mamatnabiyev

Author 2: Christos Chronis

Author 3: Iraklis Varlamis

Author 4: Meirambek Zhaparov

Keywords: Educational robots; Internet of Things; IoT Education; Arduino for Education; IoT Educational Kit

Paper 121: Improving Potato Diseases Classification Based on Custom ConvNeXtSmall and Combine with the Explanation Model

Abstract: Potatoes are short-term crops grown for harvesting tubers. It is a type of tuber that grows on roots and is the fourth most common crop after rice, wheat, and corn. Fresh potatoes can also be used in an incredible variety of dishes by baking, boiling, or frying them. Moreover, the paper, textile, wood, and pharmaceutical industries also make extensive use of potato starch. However, soil and climate pollution are highly unfavorable for potato growth and lead to a lot of diseases such as common scab, black scurf, blackleg, dry rot, and pink rot. Thus. several types of research in medicine and computers were started for the early detection, classification, and treatment of potato diseases. In this study, transfer learning and fine-tuning were applied to potato disease classification based on a custom ConvNeXtSmall model. In addition, Gradient-weighted Class Activation Mapping (i.e., Grad-CAM) is provided for visual explanation in the final result after classification. For potato illness segmentation, k-means clustering was used to enable the difference between healthy and diseased sections based on color and texture. The data was collected from numerous websites and validated by the Bangladesh Agricultural Research Institute (i.e., BARI), including six types of potato diseases and healthy images. With a Convolutional Neural Networks (i.e., CNN) model from the Keras library, our study reached the unexpected validation accuracy, test accuracy, and F1 score in seven classifications of 99.49%, 98.97%, and 98.97%, respectively. Concerning four-class classification, high accuracy values were obtained for most of the models (i.e., 100%).

Author 1: Huong Hoang Luong

Keywords: Potato disease; classification; fine-tuning; transfer learning; Convolutional Neural Network (CNN); k-means clustering; Gradient-weighted Class Activation Mapping (Grad-CAM)

Paper 122: Dynamic Task Offloading Optimization in Mobile Edge Computing Systems with Time-Varying Workloads Using Improved Particle Swarm Optimization

Abstract: Mobile edge computing (MEC) enables offloading of compute-intensive and latency-sensitive tasks from resource-constrained mobile devices to servers at the network edge. This paper considers the dynamic optimization of task offloading in multi-user multi-server MEC systems with time-varying task workloads. The arrival times and computational demands of tasks are modeled as stochastic processes. The goal is to minimize the average task delay by optimal dynamic server selection over time. A particle swarm optimization (PSO) based algorithm is proposed that makes efficient offloading decisions in each time slot based on newly arrived tasks and pending workload across servers. The PSO-based policy is shown to outperform heuristics like genetic algorithms and simulated annealing in terms of adaptability to workload fluctuations and spikes. Experiments under varying task arrival rates demonstrate PSO’s capability to dynamically optimize time-averaged delay and energy costs through joint optimization of server selection and resource allocation. The proposed techniques provide a practical and efficient dynamic load balancing mechanism for real-time MEC systems with variable workloads.

Author 1: Mohammad Asique E Rasool

Author 2: Anoop Kumar

Author 3: Asharul Islam

Keywords: Particle Swarm Optimization (PSO); Mobile Edge Computing (MEC); Multi-User Multi-Server systems; dynamic load balancing

Paper 123: On the Combination of Multi-Input and Self-Attention for Sign Language Recognition

Abstract: Sign language recognition can be considered as a branch of human action recognition. The deaf-muted community utilizes upper body gestures to convey sign language words. With the rapid development of intelligent systems based on deep learn-ing models, video-based sign language recognition models can be integrated into services and products to improve the quality of life for the deaf-muted community. However, comprehending the relationship between different words within videos is a complex and challenging task, particularly in understanding sign language actions in videos, further constraining the performance of previous methods. Recent methods have been explored to generate video annotations to address this challenge, such as creating questions and answers for images. An optimistic approach involves fine-tuning autoregressive language models trained using multi-input and self-attention mechanisms to facilitate understanding of sign language in videos. We have introduced a bidirectional transformer language model, MISA (multi-input self-attention), to enhance solutions for VideoQA (video question and answer) without relying on labeled annotations. Specifically, (1) one direction of the model generates descriptions for each frame of the video to learn from the frames and their descriptions, and (2) the other direction generates questions for each frame of the video, then integrates inference with the first aspect to produce questions that effectively identify sign language actions. Our proposed method has outperformed recent techniques in VideoQA by eliminating the need for manual labeling across various datasets, including CSL-Daily, PHOENIX14T, and PVSL (our dataset). Furthermore, it demonstrates competitive performance in low-data environments and operates under supervision.

Author 1: Nam Vu Hoai

Author 2: Thuong Vu Van

Author 3: Dat Tran Anh

Keywords: Multi-input; self-attention; deep learning models; video-based sign language; sign language recognition

Paper 124: Improving Chicken Disease Classification Based on Vision Transformer and Combine with Integrated Gradients Explanation

Abstract: Chicken diseases are an important problem in the livestock industry, affecting the health and production performance of chicken flocks worldwide. These diseases can seriously damage the health of chickens, reduce egg production, or increase mortality, causing great economic losses to farmers. Therefore, detecting and preventing diseases in chickens is a top concern in the livestock industry, to ensure the health and sustainable production of chicken flocks. In recent years, advances in machine learning techniques have shown promise in solving challenges related to image diagnosis and classification. Leveraging the power of machine learning models, we propose the ViT16 model for disease classification in chickens, demonstrating its potential in assisting healthcare professionals to diagnose chicken flocks more effectively. In this study, ViT16 demonstrated its potential and strengths when compared with 5 models in the CNN architecture and ViT32 in the ViT architecture in the task of classifying chicken disease images with an accuracy of 99.25% - 99.75% - 100% - 98.25% in four experimental scenarios with our enhanced dataset and fine-tuning. These results were generated from transfer learning and model tuning on an augmented dataset consisting of 8067 images classified into four classes: Coccidiosis, New Castle Disease, Salmonella, and Healthy. Furthermore, the Integrated Gradients explanation has an important role in increasing the transparency and understanding of the image classification model, thereby improving and optimizing model performance. The performance evaluation of each model is done through in-depth analysis, including metrics such as precision, recall, F1 score, accuracy, and confusion matrix.

Author 1: Huong Hoang Luong

Author 2: Triet Minh Nguyen

Keywords: Vision Transformer; ViT16; classification chicken disease; transfer learning; fine-tuning; image classification; integrated gradients explanation

Paper 125: Rigorous Experimental Analysis of Tabular Data Generated using TVAE and CTGAN

Abstract: Synthetic data generation research has been progressing at a rapid pace and novel methods are being designed every now and then. Earlier, statistical methods were used to learn the distributions of real data and then sample synthetic data from those distributions. Recent advances in generative models have led to more efficient modeling of complex high-dimensional datasets. Also, privacy concerns have led to the development of robust models with lesser risk of privacy breaches. Firstly, the paper presents a comprehensive survey of existing techniques for tabular data generation and evaluation matrices. Secondly, it elaborates on a comparative analysis of state-of- the-art synthetic data generation techniques, specifically CTGAN and TVAE for small, medium, and large-scale datasets with varying data distributions. It further evaluates the synthetic data using quantitative and qualitative metrics/techniques. Finally, this paper presents the outcomes and also highlights the issues and shortcomings which are still need to be addressed.

Author 1: Parul Yadav

Author 2: Manish Gaur

Author 3: Rahul Kumar Madhukar

Author 4: Gaurav Verma

Author 5: Pankaj Kumar

Author 6: Nishat Fatima

Author 7: Saqib Sarwar

Author 8: Yash Raj Dwivedi

Keywords: Synthetic data generation; tabular data generation; data privacy; conditional generative adversarial networks; variational autoencoder

Paper 126: Packet Loss Concealment Estimating Residual Errors of Forward-Backward Linear Prediction for Bone-Conducted Speech

Abstract: This study proposes a suitable model for packet loss concealment (PLC) by estimating the residual error of the linear prediction (LP) method for bone-conducted (BC) speech. Instead of conventional LP-based PLC techniques where the residual error is ignored, we employ forward-backward linear prediction (FBLP), known as the modified covariance (MC) method, by incorporating the residual error estimates. The MC method provides precise LP estimation for a short data length, reduces the numerical difficulties, and produces a stable model, whereas the conventional autocorrelation (ACR) method of LP suffers from numerical problems. The MC method has the effect of compressing the spectral dynamic range of the BC speech, which improves the numerical difficulties. Simulation results reveal that the proposed method provides excellent outcomes from some objective evaluation scores in contrast to conventional PLC techniques.

Author 1: Ohidujjaman

Author 2: Nozomiko Yasui

Author 3: Yosuke Sugiura

Author 4: Tetsuya Shimamura

Author 5: Hisanori Makinae

Keywords: Autocorrelation method; bone-conducted speech; modified covariance method; packet loss concealment; residual error

Paper 127: A Comprehensive Analysis of Network Security Attack Classification using Machine Learning Algorithms

Abstract: As internet usage and connected devices continue to proliferate, the concern for network security among individuals, businesses, and governments has intensified. Cybercriminals exploit these opportunities through various attacks, including phishing emails, malware, and DDoS attacks, leading to disruptions, data exposure, and financial losses. In response, this study investigates the effectiveness of machine learning algorithms for enhancing intrusion detection systems in network security. Our findings reveal that Random Forest demonstrates superior performance, achieving 90% accuracy and balanced precision-recall scores. KNN exhibits robust predictive capabilities, while Logistic Regression delivers commendable accuracy, precision, and recall. However, Naive Bayes exhibits slightly lower performance compared to other algorithms. The study underscores the significance of leveraging advanced machine learning techniques for accurate intrusion detection, with Random Forest emerging as a promising choice. Future research directions include refining models and exploring novel approaches to further enhance network security.

Author 1: Abdulaziz Saeed Alqahtani

Author 2: Osamah A. Altammami

Author 3: Mohd Anul Haq

Keywords: Machine learning; cyber security; intrusion detection; network security; cyber security

Paper 128: Robust Extreme Learning Machine Based on p-order Laplace Kernel-Induced Loss Function

Abstract: Since the datasets of the practical problems are usually affected by various noises and outliers, the traditional extreme learning machine (ELM) shows low prediction accuracy and significant fluctuation of prediction results when learning such datasets. In order to overcome this shortcoming, the l2 loss function is replaced by the correntropy loss function induced by the p-order Laplace kernel in the traditional ELM. Correntropy is a local similarity measure, which can reduce the impact of outliers in learning. In addition, introducing the p-order into the correntropy loss function is rewarding to bring down the sensitivity of the model to noises and outliers, and selecting the appropriate p can enhance the robustness of the model. An iterative reweighted algorithm is selected to obtain the optimal hidden layer output weight. The outliers are given smaller weights in each iteration, significantly enhancing the robustness of the model. To verify the regression prediction of the proposed model, it is compared with other methods on artificial datasets and eighteen benchmark datasets. Experimental results demonstrate that the proposed method outperforms other methods in the majority of cases.

Author 1: Liutao Luo

Author 2: Kuaini Wang

Author 3: Qiang Lin

Keywords: p-order Laplace kernel-induced loss; extreme learning machine; robustness; iterative reweighted