IJACSA Volume 14 Issue 9 - thesai.org

IJACSA Volume 14 Issue 9

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: Data Anomaly Detection in the Internet of Things: A Review of Current Trends and Research Challenges

Abstract: The Internet of Things (IoT) has revolutionized how we interact with the physical world, bringing a new era of connectivity. Billions of interconnected devices seamlessly communicate, generating an unprecedented volume of data. However, the dramatic growth of IoT applications also raises an important issue: the reliability and security of IoT data. Data anomaly detection plays a pivotal role in addressing this critical issue, allowing for identifying abnormal patterns, deviations, and malicious activities within IoT data. This paper discusses the current trends, methodologies, and challenges in data anomaly detection within the IoT domain. In this paper, we discuss the strengths and limitations of various anomaly detection techniques, such as statistical methods, machine learning algorithms, and deep learning methods. IoT data anomaly detection carries unique characteristics and challenges that must be carefully considered. We explore these intricacies, such as data heterogeneity, scalability, real-time processing, and privacy concerns. By delving into these challenges, we provide a holistic understanding of the complexity associated with IoT data anomaly detection, paving the way for more targeted and effective solutions.

Author 1: Min Yang

Author 2: Jiajie Zhang

Keywords: Internet of things; anomaly detection; security; machine learning

Paper 2: Segmentation of Motion Objects in Video Frames using Deep Learning

Abstract: The segmentation of the moving objects in the video sequences is one of the most usable series in the machine vision field, which has absorbed the consideration of researchers in the latter decades. It is a challenging task, especially when there are several motion objects in the video, and then the system needs to discover the objects that should be segmented among the trail. Therefore, in this article, we present a new method to segment several motion objects at the same time. In this work, the propagation of the credence of the confidently-estimated frames by fine-tuning the DCNN model with the other frames is the main idea. We exert a DCNN model (which is pre-trained) for the frames to estimate the class of the object; then, we gather the frames where the approximation is locally or globally reliable. In the following, we apply a collection of the frames of CE as the training set to fine-tune the pre-trained network with the existing examples in a video. Our proposed model provides acceptable results, which are better than the results of similar models. These comparisons are made in the dataset of YouTube-VOS. Also, our presented approach is applied in the dataset of DAVIS-2017 and the obtained results are better than the results of the similar works.

Author 1: Feng JIANG

Author 2: Jiao LIU

Author 3: Jiya TIAN

Keywords: Segmentation; video processing; motion objects; deep convolutional neural network (DCNN)

Paper 3: Classification of Coherence Indices Extracted from EEG Signals of Mild and Severe Autism

Abstract: Autism spectrum disorder is a debilitating neurodevelopmental illness characterized by serious impairments in communication and social skills. Due to the increasing prevalence of autism worldwide, the development of a new diagnostic approach for autism spectrum disorder is of great importance. Also, diagnosing the severity of autism is very important for clinicians in the treatment process. Therefore, in this study, we intend to classify the electroencephalogram (EEG) signals of mild and severe autism patients. Twelve patients with mild autism and twelve patients with severe autism with the age range of 10-30 years participated in the present research. Due to the difficulties of working with autism patients and recording EEG signals from these patients in the awake state, the Emotiv Epoch headset device was utilized in this work. After signal preprocessing, we calculated short-range and long-range coherence values in the frequency range of 1-45 Hz, including short and long-range intra and inter-hemispheric coherence features. Then, statistical analysis was conducted to select coherence features with statistical differences between the two groups. Multilayer perceptron (MLP) neural network and support vector machine (SVM) with radial basis function (RBF) kernel were used in the classification stage. Our results showed that the best MLP classification performance was obtained by selected inter-hemispheric coherence features with accuracy, sensitivity and specificity of 96.82%, 97.82% and 96.92%, respectively. Also, the best SVM classification performance was obtained by selected inter-hemispheric coherence features with accuracy, sensitivity and specificity of 94.70%, 93.85% and 95.55%, respectively. However, it should be noted that the MLP neural network imposes a much higher computational cost than the SVM classifier. Considering that our simple system gives promising results in diagnosing autistic patients with mild and severe severities from EEG, there is scope for further work with a larger sample size and different ages and genders.

Author 1: Lingyun Wu

Keywords: Autism spectrum disorder; electroencephalography (EEG); classification; neural network; support vector machine; coherence feature

Paper 4: Enhancing Breast Cancer Diagnosis using a Modified Elman Neural Network with Optimized Algorithm Integration

Abstract: Breast cancer is a class of cancer that starts in the cells of the breast. It happens once the cells of the breast divide and amplify abnormally and uncontrollably. Other parts of the body, including lymph nodes, bones, lungs, and liver, can be affected by breast cancer. Early diagnosis and treatment are critical in helping to lessen the risk of death from breast cancer. Machine learning is a type of artificial intelligence that can be used to diagnose breast cancer. It uses algorithms to analyze data and assess patterns associated with breast cancer. Machine learning models can help improve diagnostic accuracy, reduce false-positive results, and improve the efficiency of diagnosis. Elman Neural Networks (ENNs) are machine learning algorithms that can be used to diagnose breast cancer. ENNs use medical data to detect patterns that are associated with the presence of cancer. The accuracy of ENNs in diagnosing breast cancer is still being researched, but they have the potential to help improve diagnostic accuracy and reduce false-positive results. In the existing study, a new modified version of ENN founded on a combination of an upgraded version of the imperialist competitive algorithm is proposed for this objective. Likewise, the results of the model compared with some other methods illustrated the proposed method's higher efficiency.

Author 1: Linkai Chen

Author 2: CongZhe You

Author 3: Honghui Fan

Author 4: Hongjin Zhu

Keywords: Breast cancer model; Elman Neural Network; upgraded imperialist competitive algorithm

Paper 5: Stacked LSTM and Kernel-PCA-based Ensemble Learning for Cardiac Arrhythmia Classification

Abstract: Cardiovascular diseases (CVD) are the most prevalent causes of death and disability worldwide. Cardiac arrhythmia is one of the chronic cardiovascular diseases that create panic in human life. Early diagnosis aids physicians in securing life. ECG is a non-stationary physiological signal representing the heart's electrical activity. Automated tools to detect arrhythmia from ECG signals are possible with Machine Learning (ML). The ensemble learning technique combines the power of two or more classifiers to solve a computational intelligence problem. It enhances the performance of the models by fusing two or more models, which extremely increases its strength. The proposed ensemble Machine learning amalgamates the potency of Long Short-Term Memory (LSTM) and ensemble learning, opening up a new direction for research. In this research work, two novel ensemble methods of Extreme Gradient Boosting-LSTM (EXGB-LSTM) are developed, which use LSTM as a base learner and are transformed into an ensemble learner by coalescing with Extreme Gradient Boosting. Kernel Principal Component Analysis (K-PCA) is a significant non-linear dimensionality reduction technique. It can manage high-dimensional datasets with various features by lowering the dimensionality of the data while retaining the most crucial details. It has been applied as a preprocessing step for feature reduction in the dataset, and the performance of EXGB-LSTM is tested with and without K-PCA. Experimental results showed that the first method, fusion of EXG-LSTM, has reached an accuracy of 92.1%, Precision of 90.6%, F1-score of 94%, and Recall of 92.7%. The second proposed method, KPCA with EXGB-LSTM, attained the highest accuracy of 94.3%, with a precision of 92%, F1-score of 98%, and Recall of 94.9% for multi-class cardiac arrhythmia classification.

Author 1: Azween Abdullah

Author 2: S. Nithya

Author 3: M. Mary Shanthi Rani

Author 4: S. Vijayalakshmi

Author 5: Balamurugan Balusamy

Keywords: Arrhythmia classification; ensemble learning; extreme gradient boosting; kernel PCA; LSTM; machine learning

Paper 6: A Versatile Shuffle Resource Units Recomputation Algorithm for Uplink OFDMA Random Access

Abstract: IEEE 802.11ax introduces Uplink Orthogonal Frequency Division Multiple Access (OFDMA)-based Random Access (UORA), a novel feature for facilitating random channel access in Wireless Local Area Networks (WLANs). Similar to the conventional random access scheme in WLANs, UORA employs the OFDMA backoff (OBO) procedure to access the channel’s Resource Units (RUs) and selects a random OBO counter within the OFDMA contention window (OCW) range. The Access Point (AP) can determine and communicate this OCW range to each station (STA). Multiple STAs accessing RUs result in transmission failure due to RU collisions, which occur when specific RUs remain unassessed by any STA, leading to wastage. Efforts to optimize channel efficiency require minimizing both collisions and idle RU despite the challenges arising from UORA’s distributed and random nature. The Fisher-Yates shuffle algorithm introduces a random uniform distribution strategy for managing RU allocations among STAs. The results demonstrate that this approach enables STAs to access RUs in a distributed manner, effectively reducing idle and wasted RUs, especially in scenarios involving a limited number of STAs. Furthermore, this approach effectively mitigates collisions among STAs, even in scenarios with a more significant number of STAs.

Author 1: Azyyati Adiah Zazali

Author 2: Shamala Subramaniam

Author 3: Zuriati Ahmad Zukarnain

Author 4: Abdullah Muhammed

Keywords: IEEE 802.11ax; OFDMA; UORA; random access; backoff; resource units allocation; multi-user

Paper 7: The Promise of Self-Supervised Learning for Dental Caries

Abstract: Self-supervised learning (SSL) is a type of machine learning that does not require labeled data. Instead, SSL algorithms learn from unlabeled data by predicting the order of image patches, predicting the missing pixels in an image, or predicting the rotation of an image. SSL has been shown to be effective for a variety of tasks, including image classification, object detection, and segmentation. Dental image processing is a rapidly growing field with a wide range of applications, such as caries detection, periodontal disease progression prediction, and oral cancer detection. However, the manual annotation of dental images is time-consuming and expensive, which limits the development of dental image processing algorithms. In recent years, there has been growing interest in using SSL for dental image processing. SSL algorithms have the potential to overcome the challenges of manual annotation and to improve the accuracy of dental image analysis. This paper conducts a comparative examination between studies that have used SSL for dental caries processing and others that use machine learning methods. We also discuss the challenges and opportunities for using SSL in dental image processing. We conclude that SSL is a promising approach for dental image processing. SSL has the potential to improve the accuracy and efficiency of dental image analysis, and it can be used to overcome the challenges of manual annotation. We believe that SSL will play an increasingly important role in dental image processing in the years to come.

Author 1: Tran Quang Vinh

Author 2: Haewon Byeon

Keywords: Machine learning; dental imaging; dental caries; oral diseases

Paper 8: Optimization Method for Trajectory Data Based on Satellite Doppler Velocimetry

Abstract: Due to cost and energy consumption limitations, there are significant differences in the positioning capabilities of mobile terminals, resulting in unsatisfactory quality of trajectory data. In this paper, satellite Doppler data is used to optimize trajectory data. First, the system state equation is established by the kinematic relationship between the measured velocity and position, and the static linear Kalman filter estimates the optimal system state. Then a dynamic Kalman filter system is established by correlating the measurement error matrix parameters of the Kalman filter with the vertical dilution of precision of satellite positioning. Finally, the whole-day trajectory of a taxi in Shenzhen was visualized, and the deviation between the trajectory points and the urban road was calculated to compare the optimized and non-optimized taxi trajectories. The results show that the proposed optimization method can effectively reduce the deviation between trajectory points and urban roads, and this method can be used to process vehicle trajectory data in urban traffic research.

Author 1: Junzhuo Li

Author 2: Wenyong Li

Author 3: Guan Lian

Keywords: Urban transportation; Kalman Filter; information fusion; trajectory data

Paper 9: Optimized YOLOv7 for Small Target Detection in Aerial Images Captured by Drone

Abstract: It is challenging to detect small targets in aerial images captured by drones due to variations in target sizes and occlusions arising from the surrounding environment. This study proposes an optimized object detection algorithm based on YOLOv7 to address the above-mentioned challenges. The proposed method comprises the design of a Genetic Kmeans (1- IoU) clustering algorithm to obtain customized anchor boxes that more significantly apply to the dataset. Moreover, the SPPFCSPC_group structure is optimized using group convolutions to reduce model parameters. The fusion of Spatial Pyramid Pooling-Fast (SPPF) and Cross Stage Partial (CSP) structures leads to increased detection accuracy and enhanced multi-scale feature fusion network. Furthermore, a Detect Head is incorporated into the classification phase for more accurate position and class predictions. According to experimental findings, the optimized YOLOv7 algorithm performs quite well on the VisDrone2019 dataset in terms of detection accuracy. Compared with the original YOLOv7 algorithm, the optimized version shows a 0.18% increase in the Average Precision (AP), a reduction of 5.7 M model parameters, and a 1.12 Frames Per Second (FPS) improvement in the frame rate. With the above-described enhancements in AP and parameter reduction, the precision of small target detection and the real-time detection speed are increased notably. In general, the optimized YOLOv7 algorithm offers superior accuracy and real-time capability, thus making it well-suited for small target detection tasks in real-time drone aerial photography.

Author 1: Yanxin Liu

Author 2: Shuai Chen

Author 3: Lin Luo

Keywords: Small target detection; drone aerial photography; YOLOv7; clustering algorithm; spatial pyramid pooling

Paper 10: DevOps Implementation Challenges in the Indonesian Public Health Organization

Abstract: The importance of accelerating software development to meet rapidly changing business needs has driven the Indonesian Public Health Organization (IPHO) to adopt DevOps. But after three years, the expected benefits have not been achieved. This research aims to identify the main challenges and obstacles in implementing DevOps at IPHO. A comprehensive examination of existing literature is employed to recognize prevalent difficulties encountered by organizations when implementing DevOps. The main factors are ranked using the Fuzzy Analytic Hierarchy Process (FAHP) based on survey data from DevOps practitioners at IPHO. This study helps fill in some gaps left by empirical studies on the challenges in applying DevOps, especially in the public healthcare sector. It also streamlines the data collection and analysis process by utilizing FAHP, simplifying the survey process, and reducing the number of questions compared to previous approaches. According to the research findings, the primary hurdle that requires attention is the mindset to transform from a traditional approach to continuous delivery. In addition, the lack of understanding about the benefits of implementing DevOps and the lack of cross-functional leadership are also identified as challenges that need to be considered. However, IPHO does not view the use of legacy tools and technologies as a significant impediment to adopting DevOps.

Author 1: Muhammad Yazid Al Qahar

Author 2: Teguh Raharjo

Keywords: DevOps; challenges; fuzzy AHP; software development

Paper 11: A Bibliometric Analysis of Smart Home Acceptance by the Elderly (2004-2023)

Abstract: Both academia and business firmly endorse the notion that a smart home would be the solution to easing the excessive social burden associated with demographic ageing and improving older adults' quality of life by enhancing living independence while encouraging their desire to age in place. This study uses bibliometric analysis to examine the research trends on elderly people's acceptance of smart home. The results are derived from analysis using the VOSviewer software on 257 documents in the Scopus database. The results reveal that: there is an accelerating growth rate for the smart home literature focusing on the elderly’s acceptance since 2004; the majority of these studies are journal articles filed in the research area of computer science; the most commonly mentioned keywords include “smart home(s)” and “older adults”; the US has produced the highest number of related works; and the most cited articles are composed by authors across nations with tight collaborations.

Author 1: Bo Yuan

Author 2: Norazlyn Kamal Basha

Keywords: Smart home; acceptance; elderly people; ageing-in-place; bibliometric analysis; VOSviewer

Paper 12: Automatic Generation of Image Caption Based on Semantic Relation using Deep Visual Attention Prediction

Abstract: While modern systems for managing, retrieving, and analyzing images heavily rely on deriving semantic captions to categorize images, this task presents a considerable challenge due to the extensive capabilities required for manual processing, particularly with large images. Despite significant advancements in automatic image caption generation and human attention prediction through convolutional neural networks, there remains a need to enhance attention models in these networks through efficient multi-scale features utilization. Addressing this need, our study presents a novel image decoding model that integrates a wavelet-driven convolutional neural network with a dual-stage discrete wavelet transform, enabling the extraction of salient features within images. We utilize a wavelet-driven convolutional neural network as the encoder, coupled with a deep visual prediction model and Long Short-Term Memory as the decoder. The deep Visual Prediction Model calculates channel and location attention for visual attention features, with local features assessed by considering the spatial-contextual relationship among objects. Our primary contribution is to propose an encoder and decoder model to automatically create a semantic caption on the image based on the semantic contextual information and spatial features present in the image. Also, we improved the performance of this model, demonstrated through experiments conducted on three widely used datasets: Flickr8K, Flickr30K, and MSCOCO. The proposed approach outperformed current methods, achieving superior results in BLEU, METEOR, and GLEU scores. This research offers a significant advancement in image captioning and attention prediction models, presenting a promising direction for future work in this field.

Author 1: M. M. EL-GAYAR

Keywords: Semantic image captioning; deep visual attention model; long short-term memory; wavelet driven convolutional neural network

Paper 13: Enhancing Oil Price Forecasting Through an Intelligent Hybridized Approach

Abstract: The oil market has long experienced price fluctuations driven by diverse factors. These shifts in crude oil prices wield substantial influence over the costs of various goods and services. Moreover, the price per barrel is intricately intertwined with global economic activities, themselves influenced by the trajectory of oil prices. Analyzing oil behavior stands as a pivotal means for tracking the evolution of barrel prices and predicting future oil costs. This analytical approach significantly contributes to the field of crude oil price forecasting. Researchers and scientists alike prioritize accurate crude oil price forecasting. Yet, such endeavors are often challenged by the intricate nature of oil price behavior. Recent times have witnessed the effective employment of various approaches, including Hybrid and Machine Learning techniques to address similarly complex tasks, though they often yield elevated error rates, as observed in financial markets. In this study, the goal is to enhance the predictive precision of several weak supervised learning predictors by harnessing hybridization, particularly within the context of the crude oil market's multifaceted variations. The focus extends to a vast dataset encompassing CPSE Stock ETF prices over a period of 23 years. Ten distinct models, namely SVM, XGBoost, Random Forest, KNN, Gradient Boosting, Decision Tree, Ridge, Lasso, Elastic Net, and Neural Network, were employed to derive elemental predictions. These predictions were subsequently amalgamated via Linear Regression, yielding heightened performance. The investigation underscores the efficacy of hybridization as a strategy. Ultimately, the proposed approach's performance is juxtaposed against its individual weak predictors, with experiment results validating the findings.

Author 1: Hicham BOUSSATTA

Author 2: Marouane CHIHAB

Author 3: Younes CHIHAB

Author 4: Mohammed CHINY

Keywords: Oil market; prediction; crude oil; hybrid approach; CPSE stock ETF price; machine learning; stock markets

Paper 14: Compression Analysis of Hybrid Model Based on Scalable WDR Method and CNN for ROI-based Medical Image Transmission

Abstract: The image compression techniques are the fast-growing methods and have developed on large scale. Among them, wavelet-based compression methods are most promising and efficient techniques widely used in the field of medical image processing and transmission. The compression techniques are treated as lossy or lossless models and these can be applied on the medical images considering different situations. The medical image parts are separated into two regions. The central part of the image is treated as core region called region of interest (ROI) and others are treated as non-ROI. ROI based coding techniques are considered as most important in the medical field for efficient transmission of clinical data. The proposed method focuses on these concepts. The ROI parts considered are either smooth or textured regions. These are extracted using a segmentation method called singular value decomposition (SVD) method. An efficient run length coding method called wavelet difference reduction method (WDR) with region growing approach is used to code the extracted ROI part after applying 5/3 based integer wavelet transform. The remaining parts called non-ROI part or background artifacts are coded using Convolution Neural etwork (CNN) method. The proposed method is also restructured as layered structure to achieve adaptive scalable property and named as scalable WDR-CNN (SWDR-CNN) method. The proposed SWDR-CNN method has been evaluated using rate distortion metrics such as Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The coding gains in terms of PSNR values of SWDR-CNN method has been analysed and compared to popular scalable algorithm like S-SPIHT. The SWDR-CNN method has achieved better coding gain from 0.2 dB to 6 dB in terms of PSNR values. Hence, it is proved that proposed model can be used to code the ROI of images and has applications in the field of medical image data coding and transmission.

Author 1: Bindulal T.S

Keywords: Medical image segmentation; compression; region of interest; wavelet difference reduction; convolutional neural network; singular value decomposition

Paper 15: A Proposed Intelligent Model with Optimization Algorithm for Clustering Energy Consumption in Public Buildings

Abstract: Recently, intelligent applications gained a significant role in the energy management of public buildings due to their ability to enhance energy consumption performance. Energy management of these buildings represents a big challenge due to their unexpected energy consumption characteristics and the deficiency of design guidelines for energy efficiency and sustainability solutions. Therefore, an analysis of energy consumption patterns in public buildings becomes necessary. This reveals the significance of understanding and classifying energy consumption patterns in these buildings. This study seeks to find the optimal intelligent technique for classifying energy consumption of public buildings into levels (e.g., low, medium, and high), find the critical factors that influence energy consumption, and finally, find the scientific rules (If-Then rules) to help decision-makers for determining the energy consumption level in each building. To achieve the objectives of this study, correlation coefficient analysis was used to determine critical factors that influence on energy consumption of public buildings; two intelligent models were used to determine the number of clusters of energy consumption patterns which are Self Organizing Map (SOM) and Batch-SOM based on Principal Component Analysis (PCA). SOM outperforms Batch-SOM in terms of quantization error. The quantization error of SOM and Batch-SOM is 8.97 and 9.24, respectively. K-means with a genetic algorithm were used to predict cluster levels in each building. By analyzing cluster levels, If-Then rules have been extracted, so needs that decision-makers determine the most energy-consuming buildings. In addition, this study helps decision-makers in the energy field to rationalize the consumption of occupants of public buildings in the times that consume the most energy and change energy suppliers to those buildings.

Author 1: Ahmed Abdelaziz

Author 2: Vitor Santos

Author 3: Miguel Sales Dias

Keywords: Energy consumption in public buildings; self-organizing map; K-means; genetic algorithm; principal component analysis

Paper 16: A Survey of Evolving Performance Analysis Technologies, Algorithms and Models for Sports

Abstract: The emergence and extensive development and deployment of Industrial Revolution 4.0 have distinctly transformed the methodologies of sports performance monitoring. Consequently, there has been an increase in the emergence of new and adapted technologies in various areas of sports, such as competition analysis, player performance analysis and many others. There are rich and heterogeneous sports performance analysis technologies, algorithms and frameworks which provide constant basis for elevating new horizons of sports technologies. Thus, this paper aims to encompass significant findings that will provide a comprehensive survey in this area. Previous surveys have extensively focused on various methodologies of sports performance analysis, sport-specific analysis and other technology revolving around sports performance analysis. However, most of the focus is largely on training and competition performances and not off-field. The objective of this paper is to understand the current research trends, challenges and future directions of dynamically evolving technology embedded in the world of sports. This survey aims at contributing to this rich repository but with a new focus element of off-field that researches the connection between the athlete, the sports aspect of their life, the non-sport aspect and the methodologies of sports performance analysis. In addition, the exponential growth of Artificial Intelligence (AI) as a base for sports performance analysis systems and platforms is analysed extensively. This paper also presents a comprehensive classification of athlete performance analysis using algorithm tools and sports performance platforms and systems. Subsequently, the detailed analysis of this taxonomy has enabled the identification and detailed analysis of open issues and future directions.

Author 1: Shamala Subramaniam

Author 2: Manoj Ravi Shankar

Author 3: Azyyati Adiah Zazali

Author 4: Hong Siaw Swin

Author 5: Zarina Muhamed

Author 6: Sivakumar Rajagopal

Author 7: Mohamad Zamri Napiah

Author 8: Faisal Embung

Keywords: Sports performance analysis technology; on-field analysis; IoT; real-time monitoring; off-field analysis

Paper 17: An Improvement for Spatial-Temporal Queries of ATMGRAPH

Abstract: As a knowledge graph for the field of ATM (Air Traffic Management), ATMGRAPH integrates aviation information from various sources, and provides a new way to comprehensively analyze ATM data, but the storage schema of ATMGRAPH is inefficient for trajectory-related queries which have typical spatial-temporal characteristics, thus cannot meet the application requirements. This paper presents an improved storage model of ATMGRAPH, specifically, we design a cluster structure to connect trajectory points and spatial-temporal information to speed up trajectory-related queries, and we link flights, airports, and weather information in an effective way to speed up weather-related queries. We create a dataset of about 10,000 real domestic flights, and build a knowledge graph of it which contains about 11.66 million triplets. Experimental results show that ATM knowledge graph constructed by this storage model can significantly improve the efficiency of spatial-temporal related queries.

Author 1: ZHANG Zhiyuan

Author 2: HAN Boyang

Keywords: Air traffic management; knowledge graph; storage model; spatial-temporal query; ontology

Paper 18: Comparison of Machine Learning Algorithms for Crime Prediction in Dubai

Abstract: This study aims to find the most accurate algorithm that is capable of predicting crimes in Dubai. It compares models on a dataset of sample crimes in the Emirate of Dubai, United Arab Emirates using the open-source data mining software WEKA, which enabled us to use Random Forest, KNN, SVM, ANN, Naïve Bayes and Decision Tree, We chose those algorithms as former studies that were effective used them. We have applied the algorithms on a dataset containing 13440 Major Crime in four categories occurred between 2014 and 2018. After comparing the models and analyzing their success rates, we identified the ideal algorithms and evaluated the effectiveness of variables in making predictions by measuring the correlation coefficients. One of the study's most crucial recommendations is to increase the variables and data, also adding more details about the crime, the criminal, and the victim. These variables make an impact on the analysis and the ultimate prediction.

Author 1: Shaikha Khamis AlAbdouli

Author 2: Ahmad Falah Alomosh

Author 3: Ali Bou Nassif

Author 4: Qassim Nasir

Keywords: Machine learning; crime analysis; crime patterns; KNN; random forest; SVM; ANN; Naïve Bayes; Decision Tree; major crime

Paper 19: Preserving Cultural Heritage Through AI: Developing LeNet Architecture for Wayang Image Classification

Abstract: Wayang, an ancient cultural tradition in Java, has been an integral part of Indonesian culture for 1500 years. Rooted in Hindu cultural influences, wayang has evolved into a highly esteemed and beloved performance art. In the form of wayang kulit, this tradition conveys profound philosophical messages and implicit meanings that resonate with Javanese society. This research aims to develop an artificial intelligence (AI) model using deep learning with the LeNet architecture to accurately classify wayang images. The model was tested with 2515 Punakawan wayang images, showing excellent performance with an accuracy of 80% to 85%. Although the model successfully recognizes and distinguishes wayang classes, it faces some challenges in classifying specific classes, particularly in scenarios 2 and 4. Nevertheless, this research has a positive impact on cultural preservation, as the developed AI model can be used for automatic wayang image recognition. These implications open opportunities to better understand and preserve this rich cultural heritage through AI technology. With further improvements, this model has the potential to become a valuable tool in the efforts to preserve and introduce wayang culture to future generations.

Author 1: Muhathir

Author 2: Nurul Khairina

Author 3: Rehia Karenina Isabella Barus

Author 4: Mutammimul Ula

Author 5: Ilham Sahputra

Keywords: Wayang; LeNet; artificial intelligence; deep learning; cultural tradition

Paper 20: A Comprehensive Review of Modern Methods to Improve Diabetes Self-Care Management Systems

Abstract: Diabetes mellitus has become a global epidemic, with an increasing number of individuals affected by this chronic metabolic disorder. Effective management of diabetes requires a comprehensive self-care approach, which encompasses various aspects like monitoring blood glucose levels, adherence to medication, modifications in lifestyle, and regular healthcare monitoring. Innovative techniques for bettering diabetic self-care management have been developed recently as a result of developments in technology and healthcare systems. This comprehensive review examines the modern methods that have emerged to enhance diabetes self-care management systems. The review focuses on the integration of technology, Behavioural Change Techniques (BCTs), behavioural health theories such as Transtheoretical Model (TTM), the Health Belief Model (HBM), Theory of Reasoned Action/Planned Behaviour (TPB), Social Cognitive Theory (SCT) techniques to promote optimal diabetes care outcomes. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 standards were followed in this research's documentation. The Systematic Literature Review (SLR) period, which covered 2009 to 2020, was used to acquire the most recent complete review. Overall, the SLR results show that self-care interventions have a favourable impact on behaviours modification, the encouragement of good lifestyle habits, the lowering of blood glucose scales, and the accomplishment of significant weight loss. According to the review's findings, treatments for diabetic self-management that included behavioural health theories and BCTs in their creation tended to be more successful. In order to assist academics and practitioners with the creation of future applications, the restriction and future direction were finally defined. After recognising the potential for combining BCT methodologies and theories, it creates self-management interventions. Depending on these recognised cutting-edge mechanisms, the current SLR can assist application developers create a model to construct efficient self-care interventions for diabetes.

Author 1: Alhuseen Omar Alsayed

Author 2: Nor Azman Ismail

Author 3: Layla Hasan

Author 4: Farhat Embarak

Keywords: Diabetes self-care; diabetes management; systematic literature review; BCT theories

Paper 21: An Improved Convolutional Neural Network for Churn Analysis

Abstract: The significance of customer churn analysis has escalated due to the increasing availability of relevant data and intensifying competition. Researchers and practitioners are focused on enhancing prediction accuracy in modeling approaches, with deep neural networks emerging as appealing due to their robust performance across domains. However, the computational demands surge due to the challenges posed by dimensionality and inherent characteristics of the data. To address these issues, this research proposes a novel hybrid model that strategically integrates Convolutional Neural Networks (CNN) and a modified Variational Autoencoder (VAE). By carefully adjusting the parameters of the VAE to capture the central tendency and range of variation, the study aims to enhance the effectiveness of classifying high-dimensional churn data. The proposed framework's efficacy is evaluated using six benchmark datasets from various domains, with performance metrics encompassing accuracy, f1-score, precision, recall, and response time. Experimental results underscore the prowess of the hybrid technique in effectively handling high-dimensional and imbalanced time series data, thus offering a robust pathway for enhanced churn analysis.

Author 1: Priya Gopal

Author 2: Nazri Bin MohdNawi

Keywords: Customer churn analysis; deep learning; variational autoencoder; convolutional neural networks; dimensionality reduction

Paper 22: A New Method for Classifying Intracerebral Hemorrhage (ICH) Based on Diffusion Weighted –Magnetic Resonance Imaging (DW-MRI)

Abstract: Stroke is a condition where the blood supply to the brain is cut off. This occurs due to the rupture of blood vessels in the intracerebral area or Intracerebral Hemorrhage (ICH). Examination by health workers is generally carried out to get an overview of the part of the brain of a patient who has had a stroke. The weakness in diagnosing this disease is that deeper knowledge is needed to classify the type of stroke, especially ICH. This study aims to use the Modified Layers Convolutional Neural Network (ML-CNN) method to classify ICH stroke images based on Diffusion-Weighted (DW) MRI. The data used in this study is a DWI stroke MRI image dataset of 3,484 images. The data consists of 1,742 normal and ICH images validated by a radiologist. Because the data used is relatively small and takes into account the computational time, Stochastic Gradient Descent (SGD) is used. This study compares the basic CNN model scenario with the addition of layers to the original CNN model to produce the highest accuracy value. Furthermore, each model is cross-validated with a different k to produce performance in each model as well as changes to batch size and epoch and comparison with machine learning models such as SVM, Random Forest, Extra Trees, and kNN. The results showed that the smaller the number of batch sizes, the higher the accuracy value and the number of epochs, the higher the number of epochs, the higher the accuracy value of 99.86%. Then, four machine learning methods with accuracy, sensitivity, and specificity below 90% are all compared to CNN2. As a summary of this research, the proposed CNN modification works better than the four machine learning models in classifying stroke images.

Author 1: Andi Kurniawan Nugroho

Author 2: Jajang Edi Priyanto

Author 3: Dinar Mutiara Kusumo Nugraheni

Keywords: Batch size; Epoch; ML-CNN; SGD; Stroke

Paper 23: Application Prototype for Inclusive Literacy for People with Reading Disabilities

Abstract: This article details the process of creating a prototype mobile application that aims to promote inclusive literacy for people with reading disabilities. The goal of this application is to help people with reading difficulties to become more independent so that they can participate in society and take advantage of educational and employment opportunities that were previously unavailable to them. The methodology used in this work is Design Thinking as it is a user-centered creative approach to solving difficult challenges and addresses creativity, design and problem solving. The results obtained from the expert judgment based on Atlas TI 22 provide a valuable perspective on the viability and potential of these technological tools. The analysis of the results of the application prototype designs gives an encouraging picture of 85%. Similarly, 75% confirm that the app effectively complements inclusive literacy efforts, a significant achievement in line with the objective, and 70% appreciate the app's interaction with people with reading disabilities. Finally, a staggering 87% would gladly recommend the app, underscoring its valuable impact. In conclusion, the article discusses how mobile applications can help people with reading difficulties become more literate. The good reception of the prototype confirms the importance of technology in inclusive education and the value of this approach to improving the lives and education of this demographic.

Author 1: Laberiano Andrade-Arenas

Author 2: Roberto Santiago Bellido-García

Author 3: Pedro Molina-Velarde

Author 4: Cesar Yactayo-Arias

Keywords: Atlas TI 22; inclusive literacy; mobile applications; reading disability; design thinking

Paper 24: LAD-YOLO: A Lightweight YOLOv5 Network for Surface Defect Detection on Aluminum Profiles

Abstract: In this paper, we leverage the advantages of YOLOv5 in target detection to propose a highly accurate and lightweight network, called LAD-YOLO, for surface defect detection on aluminum profiles. The LAD-YOLO addresses the issues of computational complexity, low precision, and a large number of model parameters encountered in YOLOv5 when applied to aluminum profiles defect detection. LAD-YOLO reduces the model parameters and computation while also decreasing the model size by utilizing the ShuffleNetV2 module and depthwise separable convolution in the backbone and neck networks, respectively. Meanwhile, a lightweight structure called "Ghost_SPPFCSPC_group", which combines Cross Stage Partial Network Connection Operation, Ghost Convolution, Group Convolution and Spatial Pyramid Pooling-Fast structure, is designed. This structure is incorporated into the backbone along with the Convolutional Block Attention Module (CBAM) to achieve lightweight. Simultaneously, it enhances the model's ability to extract features of weak and small targets and improves its capability to learn information at different scales. The experimental results show that the mean Average Precision (mAP) of LAD-YOLO on aluminum profiles defect datasets reaches 96.9%, model size is 6.64MB, and Giga Floating Point Operations (GFLOPs) is 5.5. Compared with YOLOv5, YOLOV5s-MobileNetv3, and other networks, LAD-YOLO proposed in this paper has higher accuracy, fewer parameters, and lower floating-point computation.

Author 1: Dongxue Zhao

Author 2: Shenbo Liu

Author 3: Yuanhang Chen

Author 4: Da Chen

Author 5: Zhelun Hu

Author 6: Lijun Tang

Keywords: YOLOv5; ShuffleNetv2; lightweight and fast spatial pyramid pooling structure; convolutional block attention module; aluminum profiles surface defect detection

Paper 25: Improved YOLO-X Model for Tomato Disease Severity Detection using Field Dataset

Abstract: In the past decade, the field of automatic plant disease detection has undergone significant complexity. Advancements in convolutional neural network in deep learning have enabled the rapid and precise detection of ailments, facilitated the development of effective treatments and ultimately led to higher crop yields. One of the most challenging scenarios in plant disease occurs when multiple diseases manifest on a single leaf, exacerbating the difficulty of diagnosis due to overlapping symptoms. This study addresses these challenges by employing an enhanced YOLO-X model for detection tomato leaf diseases. The technique presented here enhances the Spatial Pyramid Pooling layer in order to extract valuable features from training data of various sizes more efficiently. We were able to increase the model's ability to identify a broader spectrum of disease symptoms by concatenating variables from multiple layers and varying sizes. In addition, we incorporate a large number of connections to increase the generalizability of the design. The application of an IoU-based (Intersection over Union) regression loss function increases the convergence of the network and the precision of the detection. For experimentation, we created a customized dataset consisting of 1220 tomato plant leaf images from various farms in Southern part of India, encompassing overlapping diseases and varying degrees of severity. The dataset includes images of healthy leaves as well as different severity levels of tomato leaf curl and tomato leaf mold stress on a single leaf. Our suggested improved SPP-based YOLO-X model beats the original YOLO-X model, according to experimental findings, which show an improvement in test dataset accuracy and a 73.42% mean Average Precision on field-collected dataset.

Author 1: Rajasree R

Author 2: C Beulah Christalin Latha

Keywords: Convolutional neural network; deep learning; object classification; plant disease detection; spatial pyramid pooling; YOLOX

Paper 26: He and She in Video Games: Impact of Gendera on Video Game Participation and Perspectives

Abstract: Playing video games is now considered one of the day-to-day activities of many adolescents and young people. This research studies the gender impact on video game participation and perspectives among college students in the Kingdom of Saudi Arabia (KSA). The data were collected by first conducting discussions involving four focus groups with a total of 26 participants to explore the topic. An online questionnaire was then distributed, and a total of 2,756 responses were received. The analysis of the data shows a clear impact of gender on the playing practices adopted, perceptions towards the pros and cons of video games, and the most used consoles and popular games. However, the practices and perspectives of male and female players did not differ regarding bullying in video games. The findings of this study can advance the understanding of this subject, and game developers who are targeting the KSA game market can use the results as the basis for developing games that are more suitable for the players in that country.

Author 1: Deena Alghamdi

Keywords: College students; gender differences; KSA; video games

Paper 27: Hyperparameter Tuning of Semi-Supervised Learning for Indonesian Text Annotation

Abstract: A crucial issue in sentiment analysis primarily relies on the annotation task involving data labeling. This critical step is typically performed by linguists, as the nuanced meaning of text significantly influences its contextual interpretation. If there is a large volume of data, annotation is time-consuming and financially burdensome. Addressing these challenges, a semi-supervised learning annotation (SSL) that integrates human annotator and artificial intelligence algorithms emerges as a potent solution. Building accurate SSL needs to explore the best architecture, including a combination of machine learning and mechanism. This research aims to construct semi-supervised model annotation text by tuning the parameter of the machine learning algorithm to gain the most accurate model. This study employed a Support Vector Machine and a Random Forest algorithm to build semi-supervised annotation. Grid-Search and Random-Search were employed to tune the Random Forest and Support Vector Machine parameters. The semi-supervised annotation model was applied to annotate Indonesian texts. The outcomes signify that hyperparameter-tuning enhances SSL performance, surpassing the performance achieved using default parameters. The experiment also shows that the SSL annotation using a Support Vector Machine tuned by Grid Search and Random Search is more robust than the Random Forest algorithm. Hyperparameter tuning is also robust to training data that contains many manual labeling errors by experts.

Author 1: Siti Khomsah

Author 2: Nur Heri Cahyana

Author 3: Agus Sasmito Aribowo

Keywords: Text annotation; semi-supervised; parameter-tuning; grid search; random search

Paper 28: Usability Testing of Memorable Word in Security Enhancing in e-Government and e-Financial Systems

Abstract: Most applications increase their security by adding an extra layer to the login process using two-factor authentication (2FA). In Saudi Arabia, One-Time Password (OTP), which is 2FA, is the most common method used as users log in to their accounts. However, some issues have emerged with using OTP as 2FA; these issues from previous research were investigated in the study. Also, the study proposed a new method of account authentication, which is a Memorable Word (MW). MW is the second and short password in which the user enters a certain number of characters instead of the whole password. The study conducted usability testing to compare two 2FA methods, OTP and MW. The study included 60 participants logged into a simulated website using both authentication methods. Then, all participants have to complete the questionnaire. The collected data analyses showed a favourable opinion of the MW method.

Author 1: Hanan Alotaibi

Author 2: Dania Aljeaid

Author 3: Amal Alharbi

Keywords: Security; usability testing; two factor authentication; one time password; memorable word

Paper 29: Enhanced Brain Tumor Detection and Classification in MRI Scans using Convolutional Neural Networks

Abstract: Tumor detection is one of the most critical and challenging tasks in the realm of medical image processing due to the risk of incorrect prediction and diagnosis when using human-aided categorization for cancer cell identification. Data input is an intensive process, particularly when dealing with a low-quality scan image, due to the background, contrast, noise, texture, and volume of data; when there are many input images to analyze, the task becomes more onerous. It is difficult to distinguish tumor areas from raw MRI scans because tumors pose a diverse appearance and superficially resemble normal tissues, which makes it more difficult to detect tumors. Deep learning techniques are applied in medical images to a great extent to understand tumor contours and areas with high intensities in input images. For timely diagnosis and the right treatment with less human involvement, and to interpret and enhance detection and classification accuracies this automated method is proposed. This proposed work is to identify and classify tumors on 2D MRI scans of the brain. In this work, a dataset is used, inside it, there are images with and without tumors of varied sizes, locations, and forms, with different image intensities and textures. In this paper, multi-layer Convolutional Neural Network (CNN) architectures are implemented. This shows two main experiments to assess the accuracy and performance of the model. First, five-layer CNN architecture with five layers and two different split ratios. Second, six-layer CNN architecture with two different split ratios. In addition, image pre-processing and hyper-parameter tuning were performed to improve the classification accuracy. The results show that the five-layer CNN architecture outperforms the six-layer CNN architecture. When results are compared with state-of-the-art methods, the proposed model for segmentation and classification is better because this model achieved an accuracy of 99.87 percent.

Author 1: Ruqsar Zaitoon

Author 2: Hussain Syed

Keywords: Multi-layer Convolutional Neural Networks (CNNs); MRI images; tumor segmentation and classification; deep learning; learning rate

Paper 30: Method for Hyperparameter Tuning of Image Classification with PyCaret

Abstract: A method for hyperparameter tuning of image classification with PyCaret is proposed. The application example compares 14 classification methods and confirms that Extra Trees Classifier has the best performance among them, AUC=0.978, Recall=0.879, Precision=0.969, F1=0.912, Time=0.609 bottom. The Extra Trees Classifier produces a large number of decision trees, similar to the random forest algorithm, but with random sampling of each tree and no permutation. This creates a dataset for each tree containing unique samples, and from the ensemble set of features a certain number of features are also randomly selected for each tree. The most important and unique property of the Extra Trees Classifier is that the feature split values are chosen randomly. Instead of using Gini or entropy to split the data to compute locally optimal values, the algorithm randomly selects split values. This makes the tree diverse and uncorrelated. i.e. the diversity of each tree. Therefore, it is considered that the classification performance is better than other classification methods. Parameter tuning of Extra Trees Classifier was performed, and training performance, test performance, ROC curve, accuracy rate characteristics, etc. were evaluated.

Author 1: Kohei Arai

Author 2: Jin Shimazoe

Author 3: Mariko Oda

Keywords: PyCaret; extra trees classifier; AUC; gini; entropy; feature split; ROC curve

Paper 31: A Novel Artifact Removal Strategy and Spatial Attention-based Multiscale CNN for MI Recognition

Abstract: The brain-computer interface (BCI) based on motor imagery (MI) is a promising technology aimed at assisting individuals with motor impairments in regaining their motor abilities by capturing brain signals during specific tasks. However, non-invasive electroencephalogram (EEG) signals collected using EEG caps often contain large numbers of artifacts. Automatically and effectively removing these artifacts while preserving task-related brain components is a key issue for MI de-coding. Additionally, multi-channel EEG signals encompass temporal, frequency and spatial domain features. Although deep learning has achieved better results in extracting features and de-coding motor imagery EEG (MI-EEG) signals, obtaining a high-performance network on MI that achieves optimal matching of feature extraction, thus classification algorithms is still a challenging issue. In this study, we propose a scheme that combines a novel automatic artifact removal strategy with a spatial attention-based multiscale CNN (SA-MSCNN). This work obtained independent component analysis (ICA) weights from the first subject in the dataset and used K-means clustering to determine the best feature combination, which was then applied to other subjects for artifact removal. Additionally, this work designed an SA-MSCNN which includes multiscale convolution modules capable of extracting information from multiple frequency bands, spatial attention modules weighting spatial information, and separable convolution modules reducing feature information. This work validated the performance of the proposed model using a real-world public dataset, the BCI competition IV dataset 2a. The average accuracy of the method was 79.83%. This work conducted ablation experiments to demonstrate the effectiveness of the proposed artifact removal method and SA-MSCNN network and compared the results with outstanding models and state-of-the-art (SOTA) studies. The results confirm the effectiveness of the proposed method and provide a theoretical and experimental foundation for the development of new MI-BCI systems, which is very useful in helping people with disabilities regain their independence and improve their quality of life.

Author 1: Duan Li

Author 2: Peisen Liu

Author 3: Yongquan Xia

Keywords: Motor Imagery (MI); Brain Computer Interface (BCI); EEG signal; artifact removal; spatial attention; Convolutional Neural Network (CNN)

Paper 32: SFFT-CapsNet: Stacked Fast Fourier Transform for Retina Optical Coherence Tomography Image Classification using Capsule Network

Abstract: The work of the Ophthalmologist in manually detecting specific eye related disease is challenging especially screening through large volume of dataset. Deep learning models can leverage on medical imaging like the retina Optical Coherence Tomography (OCT) image dataset to help with the classification task. As a result, many solutions have been proposed based on deep learning-based convolutional neural networks (CNNs). However, the limitations such as inability to recognize pose, the pooling operations which affect resolution of the featured maps have affected its performance in achieving the best accuracies. The study proposes a Capsule network (CapsNet) with contrast limited adaptive histogram equalization (CLAHE) and Fast Fourier transform (FFT), a method we called Stacked Fast Fourier Transform-CapsNet (SFFT-CapsNet). The SFFT was used as an enhancement layer to reduce noise in the retina OCT image. A two-block framework of three-layer convolutional capsule network each was designed. The dataset used for this study was presented by University of California San Diego (UCSD). The dataset consists of 84,495 X-Ray images categorized into four classes (NORMAL, CNV, DME, and DRUSEN). Experiment was conducted on the SFFT-CapsNet model and results were compared with baseline models for performance evaluation using accuracy, sensitivity, precision, specificity, and AUC as evaluation metrics. The evaluation results indicate that the proposed model outperformed the baseline model and state-of-the-arts models by achieving the best accuracies of 99.0%, 100%, and 99.8% on overall accuracy (OA), overall sensitivity (OS), and overall precision (OP), respectively. The result shows that the proposed method can be adopted to aid Ophthalmologist in retina disease diagnosis.

Author 1: Michael Opoku

Author 2: Benjamin Asubam Weyori

Author 3: Adebayo Felix Adekoya

Author 4: Kwabena Adu

Keywords: Capsule network; convolution neural network; medical imaging; optical coherence tomography

Paper 33: Deep Neural Network-based Detection of Road Traffic Objects from Drone-Captured Imagery Focusing on Road Regions

Abstract: This paper presents a novel deep learning approach for the detection of traffic objects from drone-based imagery, focusing predominantly on the rapid and accurate detection of vehicles within road sections. The proposed method consists of two primary components: a road segmentation module and a vehicle detection network. The former leverages a residual unit with skip-connections to effectively extract road areas, while the latter employs a modified version of the YOLOv3 architecture, tailored for high-accuracy and high-speed vehicle detection. To address the issue of data imbalance, which is a pervasive challenge in drone images, this paper utilizes a range of data augmentation techniques to improve the robustness of the proposed model. Experimental results on the UAVDT and UAVid datasets exhibit that the proposed model attains a substantial boost in accuracy and inference speed of vehicle detection in comparison to the existing methods. These findings underscore the potential of the proposed approach for real-world traffic monitoring applications, where rapid and reliable vehicle detection is paramount.

Author 1: Hoanh Nguyen

Keywords: Deep learning; drone images; vehicle detection; road segmentation; data imbalance

Paper 34: Strengthening Network Security: Evaluation of Intrusion Detection and Prevention Systems Tools in Networking Systems

Abstract: This study aims to enhance network security by comprehensively evaluating various Intrusion Detection and Prevention Systems tools in networking systems. The objectives of this research were to assess the performance of different IDPS tools in terms of computer resources utilization, Quality of Service metrics namely delay, jitter, throughput, and packet loss, and their effectiveness in countering Distributed Denial of Service attacks, specifically ICMP Flood and SYN Flood. The evaluation used popular IDPS tools, including Snort, Suricata, Zeek, OSSEC, and Honeypot Cowrie. Real attack scenarios were simulated to measure the tools performance. The results indicated CPU and RAM usage variations among the tools, with Snort and Suricata showing efficient resource utilization. Regarding QoS metrics, Snort demonstrated superior performance in delay, jitter, throughput, and packet loss mitigation for both attack types. The implication for further research lies in exploring the optimal configurations and fine-tuning of IDPS tools to achieve the best possible network security against DDoS attacks. This research provides valuable insights into selecting appropriate IDPS tools for network administrators, cybersecurity professionals, and organizations to fortify their infrastructure against evolving cyber threats.

Author 1: Wahyu Adi Prabowo

Author 2: Khusnul Fauziah

Author 3: Aufa Salsabila Nahrowi

Author 4: Muhammad Nur Faiz

Author 5: Arif Wirawan Muhammad

Keywords: IDPS; network security; computer performance; Quality of Service; DDoS attacks

Paper 35: Hybrid Local Search Algorithm for Optimization Route of Travelling Salesman Problem

Abstract: This study explores the Traveling Salesman Problem (TSP) in Medan City, North Sumatra, Indonesia, analyzing 100 geographical locations for the shortest route determination. Four heuristic algorithms—Nearest Neighbor (NN), Repetitive Nearest Neighbor (RNN), Hybrid NN, and Hybrid RNN—are investigated using RStudio software and benchmarked against various problem instances and TSPLIB data. The results reveal that algorithm performance is contingent on problem size and complexity, with hybrid methods showing promise in producing superior solutions. Statistical analysis confirms the significance of the differences between non-hybrid and hybrid methods, emphasizing the potential for hybridization to enhance solution quality. This research advances our understanding of heuristic algorithm performance in TSP problem-solving and underscores the transformative potential of hybridization strategies in optimization.

Author 1: Muhammad Khahfi Zuhanda

Author 2: Noriszura Ismail

Author 3: Rezzy Eko Caraka

Author 4: Rahmad Syah

Author 5: Prana Ugiana Gio

Keywords: Travelling Salesman Problem; heuristic algorithms; hybridization techniques algorithm performance; route optimization

Paper 36: A Systematic Literature Review of Computational Studies in Aquaponic System

Abstract: The word aquaponics means the growth of aquatic organisms as well as plants in the controlled environment. As the nutrients used for sustainable plant growth is obtained from aquatic organisms and the nutrients that are absorbed by the plants remediate the water for the aquatic life. The advancement in the computational studies plays a vital role in every field of life. The aim of the proposed study is to deeply analyze the computational studies that used IoT, AI, Machine learning and deep learning for aquaponic systems between the years 2019 to 2022. The literature survey deeply discuss the proposed methodology, comprehends the fundamental researches, tool, advantages, limitations, concepts, and results of the recent studies proposed by the researchers in context of aquaponic system. The proposed study extract 41 research articles from these libraries based on year of publication, title, methodology, citation, paper quality and abstract. These articles are collected from seven different research article libraries including Google Scholar, Worldwide Science, IEEE Xplore, Google Books, Refseek, ACM digital Library and Science Direct. This study develops a state of the art research for the next researchers to work on the loopholes of the previous researches in an efficient manner. The results of the proposed study shows that the implementation of IoT based machine learning and deep learning framework shows state of the art results for the nutrients regulation, sensing, monitoring and controlling of the aquaponic environment. It is concluded from the proposed study that there need to be develop ensemble learning model with an efficient dataset in context of aquaponic environment.

Author 1: Khaoula Taji

Author 2: Ali Sohail

Author 3: Yassine Taleb Ahmad

Author 4: Ilyas Ghanimi

Author 5: Sheeba Ilyas

Author 6: Fadoua Ghanimi

Keywords: Aquaponics; machine learning; internet of thing (IoT); message queue telemetry transport; sensors; SMART aquaculture

Paper 37: Cocoa Pods Diseases Detection by MobileNet Confluence and Classification Algorithms

Abstract: Cocoa cultivation is of immense importance to the people of Côte d'Ivoire. However, this culture is experiencing significant challenges due to diseases spread by various agents such as bacteria, viruses, and fungi, which cause considerable economic losses. Currently, the methods available to detect these cocoa diseases force farmers to seek the expertise of agronomists for visual inspections and diagnostics, a laborious and complex process. In the search for solutions, many studies have opted for using convolutional neural networks (CNNs) to identify diseases in cocoa pods. However, an essential advance is to develop hybrid approaches that combine the advantages of a CNN with sophisticated classification algorithms. This research stands out for its innovative contribution, combining MobileNetV2, a convolutional neural network architecture, with algorithms, such as Logistic Regression (LR), K Nearest Neighbors (KNN), Support Vector Machines (SVM), XGBoost, and Random Forest. The study was conducted in two distinct phases. First, each algorithm was evaluated individually, and then performance was measured when MobileNetV2 was merged with the algorithms mentioned. These hybrid approaches complement and amplify MobileNetV2's capabilities. To do so, they draw on MobileNetV2's inherent capabilities to extract key features and enhance information quality. By combining this expertise with the classification methods of these other models, hybrid approaches outperform individual techniques. Accuracy rates range from 72.4% to 86.04%.This performance amplitude underlines the effectiveness of the synergy between the extraction characteristics of MobileNetV2 and the classification skills of other algorithms.

Author 1: Diarra MAMADOU

Author 2: Kacoutchy Jean AYIKPA

Author 3: Abou Bakary BALLO

Author 4: Brou Médard KOUASSI

Keywords: Cocoa pods diseases; MobileNetV2; classification algorithms; machine learning; hybrid method

Paper 38: Wireless Capsule Endoscopy Video Summarization using Transfer Learning and Random Forests

Abstract: Wireless Capsule Endoscopy (WCE) is a diagnostic technique for identifying gastrointestinal diseases and abnormalities. Gastroenterologists face a considerable challenge when reviewing a lengthy video to identify a disease. The solution to this problem is generating an automated video summarization technique that generates the WCE Video summaries. This paper presents a Video Summarization technique that summarizes the WCE video. The proposed method uses transfer learning and a Random Forest classifier. Using a computationally light and pre-trained MobileNetV2 for feature extraction helped deliver results quickly. Managing small datasets and mitigating the overfitting risk was effectively addressed using Random Forest. The Random Forest's hyperparameters are optimized through the use of Bayesian optimization. The approach proposed has achieved an accuracy of 98.75% in disease prediction while significantly reducing the viewing time for the video summary. Furthermore, it has attained an average F-Score of 0.98, demonstrating its efficacy and reliability.

Author 1: Parminder Kaur

Author 2: Rakesh Kumar

Keywords: Bayesian optimization; capsule endoscopy; MobileNetV2; random forest classifier; transfer learning

Paper 39: Contributed Factors in Predicting Market Values of Loaned Out Players of English Premier League Clubs

Abstract: The top tier of the English football league division is occupied by the English Premier League (EPL). It has become a global phenomenon with exhilarating skills and has been one of the most-watched professional football leagues on the planet. The possibility of a player temporarily playing for a club other than the one to whom they are now contracted is known as a "loan player" in the English Premier League (EPL) hence, each player has a market value. Market value is an estimate of how much a player costs when a club wants to buy his contract from another club. The purpose of this study is to determine the factors that influence a player's market value at the conclusion of a loan period. With the Transfermarkt player transfer record dataset for the years 2004 through 2020, we use linear regression analysis. Our study found that a football player's market worth at the end of a loan period is influenced by several aspects, including market value at the beginning, goals, appearances, and total loan.

Author 1: Muhammad Daffa Arviano Putra

Author 2: Deshinta Arrova Dewi

Author 3: Wahyuningdiah Trisari Putri

Author 4: Retno Hendrowati

Author 5: Tri Basuki Kurniawan

Keywords: Data analytics; predicting market value; English Premier League; loaned out players; consumption; resource use

Paper 40: Exploring the Challenges and Impacts of Artificial Intelligence Implementation in Project Management: A Systematic Literature Review

Abstract: This paper presents a systematic literature review (SLR) investigating the challenges and impacts of implementing artificial intelligence (AI) in project management, specifically mapping them into the process groups defined in the Project Management Body of Knowledge (PMBOK). The study aims to contribute to the understanding of integrating AI in project management and provides insights into the challenges and impacts within each process group. The SLR methodology was applied, and a total of 34 scientific articles were analyzed. The results and analysis reveal the specific challenges and impacts within each process group. In the Initiating Process Group, AI tools and analysis techniques address challenges in risk assessment, cost prediction, and decision-making. The Planning process group benefits from various tools and methodologies that improve risk assessment, project selection, cost estimation, resource allocation, and decision-making. The Execution process group emphasizes the importance of advanced tools and techniques in enhancing productivity, resource utilization, cost reduction, and decision-making. The Monitoring and Controlling process group demonstrates the potential of advanced tools in achieving efficiency, cost reduction, improved quality, and informed decision-making. Lastly, the Closing process group emphasizes the importance of utilizing advanced tools to minimize waste, optimize resource utilization, reduce costs, improve quality, and project closure success. Overall, this research provides valuable insights and strategies for organizations seeking to implement AI in project management, thereby enhancing the potential for success within the PMBOK Process Group.

Author 1: Muhammad Irfan Hashfi

Author 2: Teguh Raharjo

Keywords: Artificial intelligence; project management; PMBOK process groups; challenge; impact

Paper 41: Deep Residual Convolutional Long Short-term Memory Network for Option Price Prediction Problem

Abstract: In the realm of financial markets, the precise prediction of option prices remains a cornerstone for effective portfolio management, risk mitigation, and ensuring overall market equilibrium. Traditional models, notably the Black-Scholes, often encounter challenges in comprehensively integrating the multifaceted interplay of contemporary market variables. Addressing this lacuna, this study elucidates the capabilities of a novel Deep Residual Convolution Long Short-term Memory (DR-CLSTM) network, meticulously designed to amalgamate the superior feature extraction prowess of Convolutional Neural Networks (CNNs) with the unparalleled temporal sequence discernment of Long Short-term Memory (LSTM) networks, further augmented by deep residual connections. Rigorous evaluations conducted on an expansive dataset, representative of diverse market conditions, showcased the DR-CLSTM's consistent supremacy in prediction accuracy and computational efficacy over both its traditional and deep learning contemporaries. Crucially, the integration of residual pathways accelerated training convergence rates and provided a formidable defense against the often detrimental vanishing gradient phenomenon. Consequently, this research positions the DR-CLSTM network as a pioneering and formidable contender in the arena of option price forecasting, offering substantive implications for quantitative finance scholars and practitioners alike, and hinting at its potential versatility for broader financial instrument applications and varied market scenarios.

Author 1: Artur Dossatayev

Author 2: Ainur Manapova

Author 3: Batyrkhan Omarov

Keywords: Deep learning; CNN; LSTM; prediction; option price

Paper 42: Factors and Models Influencing Value Co-Creation in the Supply Chain of Collection Resources for Library Distribution Providers Under Data Ecology

Abstract: Under the data ecology, the advancement of relevant technology and the utilization of relevant resources have provided more efficient technical services for various industries. However, with the proliferation of data resources, problems such as information pollution and data redundancy have arisen in the process of supply chain services for collection resources. For solving such problems and enhancing the collection resource supply efficiency of librarians, the study uses data mining technology combined with improved K-Means clustering algorithm to design a value co-creation model of library collection resource supply chain for librarians under data ecology. The outcomes indicate that the shortest running time of traditional K-Means algorithm is 40ms and the longest running time is 115ms in Wine dataset, and the running time of improved K-Means algorithm is stable at 59ms; the shortest running time of traditional K-Means algorithm is 26 ms and the longest running time is 58ms in Iris dataset, and the running time of improved K-Means algorithm is stable at 53 ms. The clustering accuracy of the improved K-Means algorithm in the Wine data set is 98.2%, which is 0.3% exceeding the traditional K-Means algorithm, which is 97.9%; the clustering accuracy in the Iris data set is 100%, which is 2.4% exceeding the traditional K-Means algorithm, which is 97.6%. In summary, it can be seen that the studied data ecology has a good application of the factors and models influencing the value co-creation of the supply chain of library resources for library dispensers.

Author 1: Xiaoyun Lin

Keywords: Resource supply chain; data mining; value co-creation; K-Means clustering algorithm; pavilion dispenser

Paper 43: A Flexible Manufacturing System based on Virtual Simulation Technology for Building Flexible Platforms

Abstract: Flexible manufacturing systems have become relatively mature in the industrial field, representing the most advanced research achievements in the development of the manufacturing industry. But currently, there are few resources and high costs in universities to create a system that is more practical, and it cannot meet the practical teaching requirements of students in multiple majors. In response to the above issues, this study first designed a flexible manufacturing system from the overall architecture, then introduced and integrated virtual simulation technology, and utilized multi-objective genetic algorithm for cargo location optimization to improve the work efficiency of the flexible system. The research results indicate that after 213 iterations of the proposed algorithm, the iteration curve of the total objective function value tends to be stable, and the effect of cargo location optimization is relatively ideal. At this time, the total objective function value is 142.5. In addition, as the scale expands, the corresponding number of iterations for multi-objective genetic algorithm at its maximum scale is 411.2. The application effect of virtual flexible manufacturing system in practical teaching in universities is good, and visual learning methods can better attract students' attention.

Author 1: Zhangchi Sun

Keywords: Flexible platform; virtual simulation technology; manufacturing control system; multi-objective genetic algorithm; slotting optimization

Paper 44: Enhanced Plagiarism Detection Through Advanced Natural Language Processing and E-BERT Framework of the Smith-Waterman Algorithm

Abstract: Effective detection has been extremely difficult due to plagiarism's pervasiveness throughout a variety of fields, including academia and research. Increasingly complex plagiarism detection strategies are being used by people, making traditional approaches ineffective. The assessment of plagiarism involves a comprehensive examination encompassing syntactic, lexical, semantic, and structural facets. In contrast to traditional string-matching techniques, this investigation adopts a sophisticated Natural Language Processing (NLP) framework. The preprocessing phase entails a series of intricate steps ultimately refining the raw text data. The crux of this methodology lies in the integration of two distinct metrics within the Encoder Representation from Transformers (E-BERT) approach, effectively facilitating a granular exploration of textual similarity. Within the realm of NLP, the amalgamation of Deep and Shallow approaches serves as a lens to delve into the intricate nuances of the text, uncovering underlying layers of meaning. The discerning outcomes of this research unveil the remarkable proficiency of Deep NLP in promptly identifying substantial revisions. Integral to this innovation is the novel utilization of the Waterman algorithm and an English-Spanish dictionary, which contribute to the selection of optimal attributes. Comparative evaluations against alternative models employing distinct encoding methodologies, along with logistic regression as a classifier underscore the potency of the proposed implementation. The culmination of extensive experimentation substantiates the system's prowess, boasting an impressive 99.5% accuracy rate in extracting instances of plagiarism. This research serves as a pivotal advancement in the domain of plagiarism detection, ushering in effective and sophisticated methods to combat the growing spectre of unoriginal content.

Author 1: Franciskus Antonius

Author 2: Myagmarsuren Orosoo

Author 3: Aanandha Saravanan K

Author 4: Indrajit Patra

Author 5: Prema S

Keywords: Natural language processing; encoder representation from transformers; document to vector + logistic regression

Paper 45: Comparative Study of Machine Learning Algorithms for Phishing Website Detection

Abstract: Phishing, a prevalent online threat where attackers impersonate legitimate organizations to obtain sensitive information from victims, poses a significant cybersecurity challenge. Recent advancements in phishing detection, particularly machine learning-based methods, have shown promising results in countering these malicious attacks. In this study, we developed and compared seven machine learning models, namely Logistic Regression (LR), k-Nearest Neighbors (KNN), Support Vector Machine (SVM), Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), and Gradient Boosting, to assess their efficiency in detecting phishing domains. Employing the UCI phishing domains dataset as a benchmark, we rigorously evaluated the performance of these models. Our findings indicate that the Gradient Boosting-based model, in conjunction with the Random Forest, exhibits superior performance compared to the other techniques and aligns with existing solutions in the literature. Consequently, it emerges as the most accurate and effective approach for detecting phishing domains.

Author 1: Kamal Omari

Keywords: Phishing detection; cybersecurity; machine learning; Gradient Boosting; Random Forest

Paper 46: Study of the Impact of the Internet of Things Integration on Competition Among 3PLs

Abstract: The Third-Party Logistics (3PL) industry plays an important role in modern supply chains, facilitating the efficient movement of goods and optimizing logistics operations. With the advent of advanced technologies, such as the Internet of Things (IoT), automation, artificial intelligence, and data analytics, the landscape of the 3PL industry has undergone significant transformation. With their tracking ability and real time data enabling capability, IoT technologies have gained great attention from researchers and practitioners and have been widely used in the supply chain sector. This paper employs the Cournot duopoly model within the framework of game theory to investigate the profound implications of the use of IoT technology on competition and operational strategies within the 3PL sector. In this study, we construct a Cournot duopoly model focusing on the assessment of the service level of third party logistics (3PL) within the market. We consider variables such as service level and the IoT adoption rates as crucial factors influencing the behavior of these firms. Through numerical simulations we quantify the impact of the technology on the overall profitability for both firms. Our findings have demonstrated the positive impact of integrating IoT on enhancing the profits of the 3PL firms. Additionally, the IoT adoption rates and the overall IoT integration costs play a critical role in determining market equilibrium and profit distribution.

Author 1: Kenza Izikki

Author 2: Mustapha Hlyal

Author 3: Aziz Ait Bassou

Author 4: Jamila El Alami

Keywords: Internet of things; third party logistics; game theory; cournot duopoly

Paper 47: A Framework for Predicting Academic Success using Classification Method through Filter-Based Feature Selection

Abstract: Students’ academic success is still a serious problem faced by higher education institutions worldwide. A strategy is needed to increase the students’ academic performance and prevent students from failing. The need to get early accurate information about poor academic performance is a must and could achieved by constructing a prediction model. Therefore, an effective technique is required to provide the accurate information and improve the accuracy of the prediction model. This study evaluates the filter-based feature selection especially the filter-based feature ranking techniques for predicting academic success. It provides a comparative study of filter-based feature selection techniques for determining the type of features (redundant, irrelevant, relevant) that affect the accuracy of the prediction models. Furthermore, this study proposes a novel feature selection technique based on attribute dependency for improving the performance of the prediction model through a framework. The experimental results show that the proposed technique significantly improved the accuracy of the prediction models from 2-8%, outperforming the existing techniques, and the Decision Tree classifier performs best for predicting with an accuracy score of 92.64%.

Author 1: Dafid

Author 2: Ermatita

Author 3: Samsuryadi

Keywords: Academic success; framework; filter-based feature selection; classifier; accuracy

Paper 48: A Performance Analysis of Point CNN and Mask R-CNN for Building Extraction from Multispectral LiDAR Data

Abstract: The extraction of buildings from multispectral Light Detection and Ranging (LiDAR) data holds significance in various domains such as urban planning, disaster response, and environmental monitoring. State-of-the-art deep learning models, including Point Convolutional Neural Network (Point CNN) and Mask Region-based Convolutional Neural Network (Mask R-CNN), have effectively addressed this particular task. Data and application characteristics affect model performance. This research compares multispectral LiDAR building extraction models, Point CNN and Mask R-CNN. Models are tested for accuracy, efficiency, and capacity to handle irregularly spaced point clouds using multispectral LiDAR data. Point CNN extracts buildings from multispectral LiDAR data more accurately and efficiently than Mask R-CNN. CNN-based point cloud feature extraction avoids preprocessing like voxelization, improving accuracy and processing speed over Mask R-CNN. CNNs can handle LiDAR point clouds with variable spacing. Mask R-CNN outperforms Point CNN in some cases. Mask R-CNN uses image-like data instead of point clouds, making it better at detecting and categorizing objects from different angles. The study emphasizes selecting the right deep learning model for building extraction from multispectral LiDAR data. Point CNN or Mask R-CNN for accurate building extraction depends on the application. For building extraction from multispectral LiDAR data, two approaches were compared utilizing precision, recall, and F1 score. The point-CNN model outperformed Mask R-CNN. The point-CNN model had 93.40% precision, 92.34% recall, and 92.72% F1 score. Mask R-CNN has moderate precision, recall, and F1.

Author 1: Asmaa A. Mandouh

Author 2: Mahmoud El Nokrashy O. Ali

Author 3: Mostafa H.A. Mohamed

Author 4: Lamyaa Gamal EL-Deen Taha

Author 5: Sayed A. Mohamed

Keywords: Multispectral LiDAR; Mask R-CNN; Point CNN; deep learning; building extraction

Paper 49: Securing IoT Devices in e-Health using Machine Learning Techniques

Abstract: The Internet of Things (IoT) has gained significance over the past several years and is currently one of the most important technologies. The capacity to link everyday objects, such as home appliances, medical equipment, autos, and baby monitors, to the internet via embedded devices with a minimum of human interaction has made continuous communication between people, processes, and things feasible. IoT devices have established themselves in many sectors, of which electronic health is considered the most important. The IoT environment deals with many private and sensitive health data that must be kept safe from tampering or theft. If safety precautions are not implemented, these dangers and assaults against IoT devices in the health sector might completely destroy this industry. Detecting security threats to an IoT environment requires sophisticated technology; these attacks can be identified using machine learning (ML) techniques, which can also predict snooping behavior based on unidentified patterns. In this paper, it is proposed to apply five strategies to detect attacks in network traffic based on the NF-ToN-IoT dataset. The classifiers used are Naive Bayes (NB), Random Forest (RF), Decision Tree (DT), Artificial Neural Network (ANN), and Support Vector Machine (SVM) models. These algorithms have been used instead of a centralized method to deliver compact security systems for IoT devices. The dataset was pre-processed to eliminate extraneous or missing data, and then a feature engineering approach was used to extract key features. The results obtained by applying each of the listed classifiers to a maximum classification accuracy of 98% achieved by the RF model showed our comparison to other work.

Author 1: Haifa Khaled Alanazi

Author 2: A. A. Abd El-Aziz

Author 3: Hedi Hamdi

Keywords: IoT; ML; DL; attack classification; e-health

Paper 50: A Multispectral Ariel Image Stitching using Decortification and EEG Signal Extraction Technique

Abstract: UAV Videos and other remote-sensing innovations have increased the demand for multispectral image stitching methods, which can gather data on a broad area by looking at different aspects of the same scene. For large-scale hyperspectral remote-sensing images, state-of-the-art techniques frequently have accumulating errors and high processing costs. However, this research paper aims to produce high-precision multispectral mapping with minimal spatial and spectral distortion. The stitching framework was created in the following manner: First, UAV collects the raw input data, which is then labeled as a signal using a connected component labeling strategy that correlates to each pixel or label using the EEG (Alpha, Beta, Theta, and Delta) technique. Next, the feature extraction process follows a novel decortication Hydrolysis CNN approach which extracts active and passive characteristics. Then after feature extraction, a novel chromatographic classification approach is employed for separating features without overfitting. Finally, a novel yield mapping georeferencing technique is employed for all images stitched together with proper alignment and segmented overlapping fields of view. The suggested deep learning model is an effective method for real-time mosaic image feature extraction which is faster by an average of 11.5 times compared to existing approaches as noted on the samples for experimental analysis.

Author 1: Mukul Manohar S

Author 2: K N Muralidhara

Keywords: EEG signal extraction; feature extraction; image stitching; multispectral image; UAV video

Paper 51: Texton Tri-alley Separable Feature Merging (TTSFM) Capsule Network for Brain Tumor Detection

Abstract: Brain tumors represent one of the most perilous and lethal forms of tumors in both children and adults. Early detection and treatment of such malignant disease types may reduce the mortality rate. However, manual procedures can be used to diagnose such disorders, and this process necessitates a careful, in-depth analysis which is prone to errors, tedious for health professionals, and time-consuming. Therefore, this research aims to design a Texton Tri-alley Separable Feature Merging (TSFM) Capsule Network based on dynamic routing, suitable for the automatic detection of brain tumors. The TSFM Capsule Network’s Texton layer helps to extract important features from the input image, and the separable convolutions coupled with the use of fewer filters and kernel sizes help to reduce the time for training, the size of the model on disk, and the number of trainable parameters generated by the model. The model’s evaluation results on the brain tumor dataset consisting of four classes show better performance than the traditional capsule network, and are comparable to the state-of-the-art models, with an overall accuracy of 97.64%, specificity of 99.24%, precision of 97.43%, sensitivity of 97.45%, f1-score of 97.44%, ROC rate of 99.50%, PR rate of 99.00%. The components and properties of the proposed model make the model deployable on devices with low memory like mobile devices. This model with better performance can assist physicians in the diagnosis of brain tumors.

Author 1: Vivian Akoto-Adjepong

Author 2: Obed Appiah

Author 3: Peter Appiahene

Author 4: Patrick Kwabena Mensah

Keywords: Texton; separable convolutions; capsule neural network; dynamic routing; brain tumor; brain tumor detection

Paper 52: Unraveling Ransomware: Detecting Threats with Advanced Machine Learning Algorithms

Abstract: In our contemporary world, the pervasive influence of information technology, computer engineering, and the Internet has undeniably catalyzed innovation, fostering unparalleled economic growth and revolutionizing education. This technological juggernaut, however, has unwittingly ushered in a parallel era of new criminal frontiers, a magnet for hackers and cybercriminals. These malevolent actors exploit the vast expanse of electronic devices and interconnected networks to perpetrate an array of cybercrimes, and among these insidious digital threats, ransomware reigns supreme. Ransomware, characterized by its ominous ability to encrypt victims' data and extort payment for its release, stands as a dire menace to individuals and organizations alike. Operating with stealth and propagating with alarming alacrity through digital networks, ransomware has emerged as a formidable adversary in the digital age. This research paper focuses on the evolving stages of ransomware, driven by cutting-edge technologies, and proposes essential methods and ideas to detect and combat this menace. The proposed methodology, anchored in Cuckoo Sandbox, PE file feature extraction, and YARA rules, orchestrates three crucial phases: data collection, feature selection, and data preprocessing, all harmonizing to strengthen our defense against this concealed cyber menace. This paper contributes to the development of effective solutions for detecting and mitigating this hidden and insidious cyber threat. This work involves the application of multiple machine learning algorithms, including LSTM, which achieves an impressive accuracy of 99% in identifying ransomware attacks.

Author 1: Karam Hammadeh

Author 2: M. Kavitha

Keywords: Ransomware; cuckoo sandbox; PEFile; YARA rules; machine learning; LSTM

Paper 53: K-Means Extensions for Clustering Categorical Data on Concept Lattice

Abstract: Formal Concept Analysis (FCA) is a key tool in knowledge discovery, representing data relationships through concept lattices. However, the complexity of these lattices often hinders interpretation, prompting the need for innovative solutions. In this context, the study proposes clustering formal concepts within a concept lattice, ultimately aiming to minimize lattice size. To address this, The study introduces introduce two novel extensions of the k-means algorithm to handle categorical data efficiently, a crucial aspect of the FCA framework. These extensions, namely K-means Dijkstra on Lattice (KDL) and K-means Vector on Lattice (KVL), are designed to minimize the concept lattice size. However, the current study focuses on introducing and refining these new methods, laying the groundwork for our future goal of lattice size reduction. The KDL utilizes FCA to build a graph of formal concepts and their relationships, applying a modified Dijkstra algorithm for distance measurement, thus replacing the Euclidean distance in traditional k-means. The defined centroids are formal concepts with minimal intracluster distances, enabling effective categorical data clustering. In contrast, the KVL extension transforms formal concepts into numerical vectors to leverage the scalability offered by traditional k-means, potentially at the cost of clustering quality due to oversight of the data's inherent hierarchy. After rigorous testing, KDL and KVL proved robust in managing categorical data. The introduction and demonstration of these novel techniques lay the groundwork for future research, marking a significant stride toward addressing current challenges in categorical data clustering within the FCA framework.

Author 1: Mohammed Alwersh

Author 2: László Kovács

Keywords: Clustering algorithms; categorical data; k-means; cluster analysis; formal concept analysis; concept lattice

Paper 54: Intelligent Heart Disease Prediction System with Applications in Jordanian Hospitals

Abstract: Heart disease is the leading cause of mortality worldwide. Early identification and prediction can play a crucial role in preventing and treating it. Based on patient data, machine learning techniques may be used to construct cardiac disease prediction models. This work aims to investigate the usage of machine learning models for heart disease prediction utilizing a publicly available dataset. The dataset contains patient information on clinical and demographic characteristics and the presence or absence of cardiac disease. Based on classification performance, many machine learning methods were tested and compared. The findings reveal that machine learning models can predict cardiac disease with accuracy and AUC values. Furthermore, the developed system is used to examine some Jordanian patients, and the predictions of the results are satisfactory. The study's findings might have far-reaching consequences for the early identification and prevention of heart disease, as well as for improving patient outcomes and lowering healthcare expenditures.

Author 1: Mohammad Subhi Al-Batah

Author 2: Mowafaq Salem Alzboon

Author 3: Raed Alazaidah

Keywords: Heart disease; machine learning; predictive models; classification; clinical data; predictions

Paper 55: A Novel Approach for Content-based Image Retrieval System using Logical AND and OR Operations

Abstract: This paper proposes an innovative ensemble learning framework for classifying medical images using Support Vector Machine (SVM) and Fuzzy Logic classifiers. The proposed approach utilizes logical AND and OR operations to combine the predictions from the two classifiers, aiming to capitalize on the strengths of each. The SVM and Fuzzy Logic classifiers were independently trained on a comprehensive database of medical images comprising various types of X-ray images. The logical OR operation was then used to create an ensemble classifier that outputs a positive classification if either of the individual classifiers does so. On the other hand, the logical AND operation was used to construct an ensemble classifier that outputs a positive classification only if both individual classifiers do so. The proposed method aims to increase sensitivity and precision by capturing as many positive instances as possible, thereby reducing false positives. The scope of the proposed work is validated in terms of overall time complexity and retrieval accuracy. The simulation outcome shows promising result with 98.36 accuracy score and 1.8 seconds to retrieve all the images in query database.

Author 1: Ranjana Battur

Author 2: Jagadisha Narayana

Keywords: Medical images; support vector machine; fuzzy logic; X-ray images; time complexity

Paper 56: Imperative Role of Digital Twin in the Management of Hospitality Services

Abstract: Digital twin implementation enables more effective terms of evaluation and planning, and also effective utilization of resources with a flood of knowledge to improve the real-time services. The hospitality industry settings utilize digital twin technologies to introduce new ideas with sensor, actuators, AR/VR improve production, and improve customer services. Currently, the hospitality industry is focused to create a fast, virtual world space where customers can get a real world of hospitality. The technologically digital twin of a vast inn office can be implemented to create both discrete and continuous event recreations in order to precisely conceptualize the events that occur in distinct frameworks. Based on the above facts, the adoption of the digital twin in the hospitality industry has gained significant attention. With this motivation, the study aims to investigate the significance and application of the digital twins in the hospitality industry for establishing innovative and digital infrastructure. In addition to this, the study discusses different elements that are significant for the digital twin. Finally, the article summarizes and recommends vital recommendation in the adoption of digital twin in hospitality industry.

Author 1: Ramnarayan

Author 2: Rajesh Singh

Author 3: Anita Gehlot

Author 4: Kapil Joshi

Author 5: Ashraf Osman Ibrahim

Author 6: Anas W. Abulfaraj

Author 7: Faisal Binzagr

Author 8: Salil Bharany

Keywords: Hospitality industry; digital twin; sensor and actuator; IoT; augment and virtual reality

Paper 57: Design of a Hypermodel using Transfer Learning to Detect DDoS Attacks in the Cloud Security

Abstract: The present research proposes a detective approach to analyzing the performance of various algorithms used for more accurate detection of Distributed Denial-of-Service (DDoS) attacks in cloud computing. From the start, this study uses machine learning and deep learning to explore whether information security has evolved in recent years. The deployment of intrusion detection systems and distributed denial-of-service attacks are then discussed. The most common DDoS attack types were summarized. In addition, this study reviewed the existing approaches and techniques for DDoS attack detection. Various pre-processing subsystems as well as attribute-based selection techniques for preventing the detection of DDoS were briefly described. The proposed Intrusion detection system uses transfer learning for detecting DDoS attacks in the Networks. The proposed system used for the data set for the Network Intrusion Detection System is SDN Dataset which has more features and is suitable to use to detect in Network Intrusions. It contains 23 features that are used to detect Intrusions in the network SDN Dataset which consists of training and testing data to detect the attacks in the network. The detection and prevention subsystems through ML and DL strategies were briefly discussed. The proposed deep learning model for DDoS attack detection in cloud storage applications is explained. After that, various preprocessing strategies employed in the detection are described, among them rebalancing data, data cleaning, data splitting, and data normalization like min-max normalization. The author created a hypermodel that consists the parameters of baseline classifiers like Support Vector Machine, K-Nearest Neighbors Algorithm, XGboost, and other various machine learning models. The proposed model gives very good accuracy compared to other machine learning models.

Author 1: Marram Amitha

Author 2: Muktevi Srivenkatesh

Keywords: Machine learning; deep learning; support vector machine; k-nearest neighbors algorithm

Paper 58: Cyberbullying Detection Based on Hybrid Ensemble Method using Deep Learning Technique in Bangla Dataset

Abstract: Globalization is certainly a blessing for us. Still, this term also brought such things that are constantly not only creating social insecurities but also diminishing our mental health, and one of them is Cyberbullying. Cyberbullying is not only a misuse of technology but also encourages social harassment among people. Research on Cyberbullying detection has gained increasing attention nowadays in many languages, including Bengali. However, the amount of work on the Bengali language compared to others is insignificant. Here we introduce a Hybrid ensemble method using a voting classifier in Bangla Cyberbullying detection and compare this with traditional Machine Learning and Deep Learning Classifiers. Before implementation, Exploratory Data Analysis was performed on the dataset to gather better insight. There are lots of papers that have already been published in other languages where it is seen that the hybrid approach provides better outcomes compared to traditional methods. Thus, we propose a highly well-driven method for Cyberbullying detection on the Bangla dataset using the hybrid ensemble method by voting classifier. The overall deployment consists of three Machine Learning classifiers, three Deep Learning classifiers, and a Hybrid approach using the voting classifier. Finally, the Hybrid ensemble method yields the best performance with an accuracy of 85%, compared with other Machine and Deep Learning methods.

Author 1: Md. Tofael Ahmed

Author 2: Afroza Sharmin Urmi

Author 3: Maqsudur Rahman

Author 4: Abu Zafor Muhammad Touhidul Islam

Author 5: Dipankar Das

Author 6: Md. Golam Rashed

Keywords: Bangla dataset; cyberbullying; exploratory data analysis; machine learning; deep learning; hybrid ensemble method

Paper 59: SE-RESNET: Monkeypox Detection Model

Abstract: The monkeypox virus, a species of the Orthopoxvirus genus within the family Poxviridae, is answerable for inflicting monkeypox. The symptoms of monkeypox last for about two to three weeks, which is often a self-limiting infection. There may be extreme cases. Recently, the case fatality rate has been in the region of 3-6. When developing a clinical medical diagnosis, it is vital to incorporate different rash diseases such as pox, measles, bacterial skin infections, scabies, syphilis, and medically connected allergies. Pathology at the symptom stage of the sickness could aid in distinctive monkeypox from chickenpox or smallpox. The dataset’s machine learning model should not be used for clinical diagnosis, but rather for developing a new model to identify illness fast. The gray scale versions of the original photos in the Monkeypox grey file could make it easier to figure out training more quickly. The channel-wise feature responses that are adaptively re-calibrated are handled by the “Squeeze-and-Excitation” (SE) block. To do this, cross-channel dependency must be explicitly modeled. To demonstrate how these architectures are put together and how these building pieces may be layered to produce SE-Resnet designs in monkeypox image sets that generalize very well. Also, demonstrate that employing SE blocks significantly enhances the performance of current state-of-the-art CNNs while incurring just a little computational cost.

Author 1: Krishnan Thiruppathi

Author 2: Selvakumar K

Author 3: Vairachilai Shenbagavel

Keywords: Squeeze-and-Excitation (SE); monkeypox; poxviridae; prodromal; chickenpox; prodromal

Paper 60: Enhancing Skin Cancer Detection Through an AI-Powered Framework by Integrating African Vulture Optimization with GAN-based Bi-LSTM Architecture

Abstract: One of the more prevalent and severe cancer kinds is thought to be skin cancer. The main objective is to detect the melanoma in initial stage and save millions of lives. One of the most difficult aspects of developing an effective automatic classification system is due to lack of large datasets. The data imbalance and overfitting problem degrades the accuracy. In this proposed work, this problem can be solved using a Generative Adversarial Network (GAN) by generating more training images. Traditional RNNs are concerned with overcoming memory constraints. By using a cyclic link on the hidden layer, these models attain Long short-term memory. However, RNNs suffer from the issue of the gradient disappearing, which affects learning performance. To overcome these challenges this work proposes Bidirectional Long Short-Term Memory (Bi-LSTM) deep learning framework for skin cancer detection. The dataset which is collected from the International Skin Imaging Collaboration were used in image processing. A novel metaheuristic enthused by the routine of African vultures is proposed in this proposed work. The African Vulture Optimisation Algorithm (AVOA) algorithm is designed to select optimum feature of skin image. The accuracy of the proposed method obtains 98.5%. This comprehensive framework, encompassing GAN-generated data, Bi-LSTM architecture, and AVOA-based feature optimization, contributes significantly to enhancing early melanoma detection.

Author 1: N. V. Rajasekhar Reddy

Author 2: Araddhana Arvind Deshmukh

Author 3: Vuda Sreenivasa Rao

Author 4: Sanjiv Rao Godla

Author 5: Yousef A.Baker El-Ebiary

Author 6: Liz Maribel Robladillo Bravo

Author 7: R. Manikandan

Keywords: Skin cancer; generative adversarial network; Bi-LSTM; African Vulture Optimisation (AVO); deep learning (DL)

Paper 61: Enhancing Diabetic Retinopathy Detection Through Machine Learning with Restricted Boltzmann Machines

Abstract: Diabetes is a potentially sight-threatening condition that can lead to blindness if left undetected. Timely diagnosis of diabetic retinopathy, a persistent eye ailment, is critical to prevent irreversible vision loss. However, the traditional method of diagnosing diabetic retinopathy through retinal testing by ophthalmologists is labor-intensive and time-consuming. Additionally, early identification of glaucoma, indicated by the Cup-to-Disc Ratio (CDR), is vital to prevent vision impairment, yet its subtle initial symptoms make timely detection challenging. This research addresses these diagnostic challenges by leveraging machine learning and deep learning techniques. In particular, the study introduces the application of Restricted Boltzmann Machines (RBM) to the domain. By extracting and analyzing multiple features from retinal images, the proposed model aims to accurately categorize anomalies and automate the diagnostic process. The investigation further advances with the utilization of a U-network model for optic segmentation and employs the Squirrel Search Algorithm (SSA) to fine-tune RBM hyperparameters for optimal performance. The experimental evaluation conducted on the RIM-ONE DL dataset demonstrates the efficacy of the proposed methodology. A comprehensive comparison of results against previous prediction models is carried out, assessing accuracy, cross-validation, and Receiver Operating Characteristic (ROC) metrics. Remarkably, the proposed model achieves an accuracy value of 99.2% on the RIM-ONE DL dataset. By bridging the gap between automated diagnosis and ophthalmological practice, this research contributes significantly to the medical field. The model's robust performance and superior accuracy offer a promising avenue to support healthcare professionals in enhancing their decision-making processes, ultimately improving the quality of care for patients with retinal anomalies.

Author 1: Venkateswara Rao Naramala

Author 2: B. Anjanee Kumar

Author 3: Vuda Sreenivasa Rao

Author 4: Annapurna Mishra

Author 5: Shaikh Abdul Hannan

Author 6: Yousef A.Baker El-Ebiary

Author 7: R. Manikandan

Keywords: Optic disc (OD); Optic cup (OC); U-network; restricted Boltzmann machines; squirrel search algorithm

Paper 62: Feline Wolf Net: A Hybrid Lion-Grey Wolf Optimization Deep Learning Model for Ovarian Cancer Detection

Abstract: Ovarian cancer is a major cause of mortality among gynecological malignancies, emphasizing the critical role of early detection in improving patient outcomes. This paper presents an automated computer-aided design system that combines deep learning techniques with an optimization mechanism for accurate ovarian cancer detection that utilizes pelvic CT images dataset. The key contribution of this work is the development of an optimized Bi-directional Long Short-Term Memory (Bi-LSTM) model which is introduced in the layers of CNN (Convolutional Neural Network), enhancing the learning process. Additionally, a feature selection method based on Lion with Grey Wolf Optimization (LGWO) is employed to enhance classifier efficiency and accuracy. The proposed approach classifies ovarian tumors as benign or malignant using the Bi-LSTM model, evaluated on the Ovarian Cancer University of Kaggle dataset. Results showcase the effectiveness of the method, achieving remarkable performance metrics, including 98% accuracy, 99.7% recall, 93% precision, and an impressive F1 score of 98%. The proposed method's efficiency is validated through comparison with validating data, demonstrating consistent and reliable results. The study's significance lies in its potential to provide an accurate and efficient solution for early ovarian cancer detection. By leveraging deep learning and optimization, the proposed method outperforms existing approaches, highlighting the promise of advanced computational techniques in improving healthcare outcomes. The findings contribute to the field of ovarian cancer detection, emphasizing the value of integrating cutting-edge technologies for effective medical diagnosis.

Author 1: Moresh Mukhedkar

Author 2: Divya Rohatgi

Author 3: Veera Ankalu Vuyyuru

Author 4: K V S S Ramakrishna

Author 5: Yousef A.Baker El-Ebiary

Author 6: V. Antony Asir Daniel

Keywords: Ovarian cancer; deep learning; bidirectional long short term memory; CT images; convolutional neural network; lion grey wolf optimization

Paper 63: Utilizing Deep Convolutional Neural Networks and Non-Negative Matrix Factorization for Multi-Modal Image Fusion

Abstract: A key element of contemporary computer vision, image fusion tries to improve the quality and interpretability of images by combining complimentary data from several image sources or modalities. This paper offers a unique method for multi-modal image fusion, combining the benefits of Deep Convolutional Neural Networks (CNNs) and Non-Negative Matrix Factorization (NMF), by using current developments in deep learning and matrix factorization techniques. Deep CNNs have shown to be remarkably effective in extracting features from images, capturing complex patterns and discriminative data. A group of deep CNNs are trained using this suggested technique on a varied dataset of multi-modal images. With the help of these networks, which extract and encode pertinent characteristics from several modalities, information-rich representations may then be combined. Concatenating, the features that were derived from the CNNs throughout the fusion process results in a fused feature representation that perfectly expresses the input modalities. The main novelty is the two-stage integration of NMF: first, breaking down the fused feature representation into non-negative basis vectors and coefficients, and then, using NMF to further extract important patterns from the fused feature maps. The non-negativity requirement in NMF guarantees the preservation of the natural structures and characteristics present in the source images, resulting in fused images that are both aesthetically pleasing and semantically intelligible. Visual examination of the merged images demonstrates the method's capacity to successfully extract important information from several modalities. The better performance and robustness of the suggested approach, which has an accuracy of roughly 99.12%, are highlighted by comparison with existing fusion approaches.

Author 1: Nripendra Narayan Das

Author 2: Santhakumar Govindasamy

Author 3: Sanjiv Rao Godla

Author 4: Yousef A.Baker El-Ebiary

Author 5: E.Thenmozhi

Keywords: Image fusion; deep convolution network; non-negative matrix factorization; multi-modal images; vector space model

Paper 64: Hybrid Image Encryption using Non-Adjacent Bits Dynamic Encoding DNA with RSA and Chaotic Systems

Abstract: Image encryption is a crucial aspect that helps to maintain the images' confidentiality and security in diverse applications. Ongoing research is focused on improving the efficiency and effectiveness of encryption. Image encryption has many practical applications in today's digital world, such as securing confidential images transmitted over networks, protecting sensitive personal information stored in images, and ensuring the privacy of medical images. The suggested work represents a breakthrough in image encryption by proposing a model that leverages the power of DNA, RSA, and chaos. This model has three phases: key generation, confusion, and diffusion. The key generation phase employs a hash function and hyperchaotic technique to generate a strong key. During the confusion phase, the positions of pixels are rearranged, either at the image level or within blocks, using the Duffing chaotic map. Once the scrambling level is determined, each pixel undergoes two successive scrambling steps, with Henon and Arnold's chaotic map to change its location. During the diffusion phase, the encryption model employs a two way approach to ensure maximum security. Firstly, it utilizes dynamic DNA cryptography for non-adjacent bits, followed by robust RSA cryptography. The experimental results indicate that the model possesses a strong security level randomness and can withstand different attacks.

Author 1: Marwa A. Elmenyawi

Author 2: Nada M. Abdel Aziem

Keywords: Cryptography; image encryption; hash function; chaotic map; DNA encoding; DNA operations; RSA algorithm

Paper 65: Object Detection and Recognition in Remote Sensing Images by Employing a Hybrid Generative Adversarial Networks and Convolutional Neural Networks

Abstract: Due to diverse backdrops, scale fluctuations, and a lack of annotated training data, the identification and recognition of objects in remote sensing images present major problems. In order to overcome these difficulties, this work suggests a novel hybrid technique that blends GAN and CNN. The suggested approach expands the small labelled dataset by synthesising realistic training examples using the generative abilities of GANs. The samples generated capture the various variances and backgrounds found in remote sensing photos, improving the object identification and recognition model's capacity to generalise. Additionally, CNNs, which are recognised for their outstanding feature extraction skills, are incorporated into the hybrid approach, enabling precise and reliable object identification and recognition. The model's CNN component is developed using both real and synthetic data, effectively combining the advantages of both fields. Several experiments are conducted on a large dataset of satellite photos to evaluate the performance of the proposed method. The results demonstrate that the hybrid model, with accuracy 97.32%, outperforms traditional approaches and pure CNN-based approaches in terms of dependability and resilience. The model may be efficiently generalised to unknown remote sensing images thanks to the GAN-generated samples, which bridge the gap among synthetic and actual data. The hybrid methodology used in this study demonstrates the possibility of merging GANs and CNNs for item detection and recognition using deep learning in remote sensing images.

Author 1: Araddhana Arvind Deshmukh

Author 2: Mamta Kumari

Author 3: V.V. Jaya Rama Krishnaiah

Author 4: Suraj Bandhekar

Author 5: R. Dharani

Keywords: Object detection; Generative Adversarial Networks (GAN); Convolutional Neural Networks (CNN); deep learning; remote sensing; satellite images; hybrid model

Paper 66: AIRA-ML: Auto Insurance Risk Assessment-Machine Learning Model using Resampling Methods

Abstract: Predicting underwriting risk has become a major challenge due to the imbalanced datasets in the field. A real-world imbalanced dataset is used in this work with 12 variables in 30144 cases, where most of the cases were classified as "accepting the insurance request", while a small percentage classified as "refusing insurance". This work developed 55 machine learning (ML) models to predict whether or not to renew policies. The models were developed using the original dataset and four data-level approaches resampling techniques: random oversampling, SMOTE, random undersampling, and hybrid methods with 11 ML algorithms to address the issue of imbalanced data (11 ML× (4 resampling techniques + unbalanced datasets) = 55 ML models). Seven classifier efficiency measures were used to evaluate these 55 models that were developed using 11 ML algorithms: logistic regression (LR), random forest (RF), artificial neural network (ANN), multilayer perceptron (MLP), support vector machine (SVM), naive Bayes (NB), decision tree (DT), XGBoost, k-nearest neighbors (KNN), stochastic gradient boosting (SGB), and AdaBoost. The seven classifier efficiency measures namely are accuracy, sensitivity, specificity, AUC, precision, F1-measure, and kappa. CRISP-DM methodology is utilisied to ensure that studies are conducted in a rigorous and systematic manner. Additionally, RapidMiner software was used to apply the algorithms and analyze the data, which highlighted the potential of ML to improve the accuracy of risk assessment in insurance underwriting. The results showed that all ML classifiers became more effective when using resampling strategies; where Hybrid resampling methods improved the performance of machine learning models on imbalanced data with an accuracy of 0.9967 and kappa statistics of 0.992 for the RF classifier.

Author 1: Ahmed Shawky Elbhrawy

Author 2: Mohamed A. Belal

Author 3: Mohamed Sameh Hassanein

Keywords: Risk assessment; machine learning; imbalanced data; rapid miner; CRISP-DM methodology

Paper 67: An MILP-based Lexicographic Approach for Robust Selective Full Truckload Vehicle Routing Problem

Abstract: Full truckload (FTL) shipment is one of the largest trucking modes. It is an essential part of the transportation industry, where the carriers are required to move FTL transportation demands (orders) at a minimal cost between pairs of locations using a certain number of trucks available at the depots. The drivers who pick up and deliver these orders must return to their home depots within a given time. In practice, satisfying those orders within a given time frame (e.g., one day) could be impossible while adhering to all operational constraints. As a result, the investigated problem is distinguished by the selective aspect, in which only a subset of transportation demands is serviced. Furthermore, travel times between nodes can be uncertain and vary depending on various possible scenarios. The robustness subsequently consists of identifying a feasible solution in all scenarios. Therefore, this study introduces an MILP-based lexicographic approach to solve a robust selective full truckload vehicle routing problem (RSFTVRP). We demonstrated the proposed method’s efficiency through experimental results on newly generated instances for the considered problem.

Author 1: Karim EL Bouyahyiouy

Author 2: Anouar Annouch

Author 3: Adil Bellabdaoui

Keywords: Vehicle routing problem; full truckload; robust optimization; MILP-based lexicographic approach; uncertain travel time

Paper 68: Design of Personalized Recommendation and Sharing Management System for Science and Technology Achievements based on WEBSOCKET Technology

Abstract: Scientific research is becoming more and more crucial to contemporary society as the backbone of the nation's innovation-driven development. The rapid growth of information technology and the rise of information technology in scientific research both contribute to the globalization of scientific research. Small research groups still don't have a place to showcase and share their accomplishments, though. In order to integrate scientific research information and combine personalised recommendation technology to suggest developments of interest to users through their historical behaviour data, the study proposes a personalised recommendation and sharing management system for scientific and technological achievements based on the Ruby on Rails framework. According to the testing results, the system had a 299ms request response time, a maximum 1KB request resource size, and a 20ms data transfer time. Additionally, the study's user-based collaborative filtering recommendation algorithm has an accuracy rate of 41% when the nearest neighbor parameter is set to 50, there are 10 information suggestions, and there are 0.7 training sets, which essentially satisfies the system criteria. In conclusion, the research suggested that a personalised recommendation and sharing management system for scientific and technological accomplishments can essentially satisfy the needs of small research teams to communicate and share scientific accomplishments, as well as realise the sharing of scientific achievements.

Author 1: Shan Zuo

Author 2: Kai Xiao

Author 3: Taitian Mao

Keywords: Research management; personalised recommendations; WebSocket; ruby on rails; informatization

Paper 69: Mechatronics Design and Development of T-EVA: Bio-Sensorized Space System for Astronaut’s Upper Body Temperature Monitoring During Extravehicular Activities on the Moon and Mars

Abstract: The exploration of the universe is progressively increasing, within this inquiry, the planet Mars and the Moon remain a mystery and challenge, as well as its colonization and civilization. Thus, in the extravehicular activities (EVA) where the astronaut will be in extreme environments performing activities such as exploration, and collection of rock and soil samples for later analysis, it should be noted that when he performs these activities, he will be exposed to extreme environmental parameters such as radiation, temperature, gravity, and many other extreme conditions. Therefore, the Center of Space Emerging Technologies (C-SET) proposed a project called T-EVA, developed into the Research Line: Space Suits and Assistive Devices, and in the Research Area: Biomechatronics and Life Support Systems, with the aim of astronaut temperature monitoring during their work outside the base station, being able to know how much their body is measuring and if they are at risk of hypothermia or hyperthermia, which could cause irreparable damage. The electronic design was made for testing both in the laboratory and outside, as well as the implementation of the lycra to mount the design, resulting in a feasible prototype that can be implemented in real situations with easy access to temperature reports.

Author 1: Paul Palacios

Author 2: Jose Cornejo

Author 3: Juan C. Chavez

Author 4: Carlos Cornejo

Author 5: Jorge Cornejo

Author 6: Mariela Vargas

Author 7: Natalia I. Vargas-Cuentas

Author 8: Avid Roman-Gonzalez

Author 9: Julio Valdivia-Silva

Keywords: Extravehicular-activities astronauts; spacesuits; body temperature; Mars; space

Paper 70: Artificial Intelligence-based Volleyball Target Detection and Behavior Recognition Method

Abstract: Volleyball has limitations in relying on judges’ subjective judgments alone to call penalties for infractions in the court. While video detail enhancement technology is extremely useful for target tracking and extraction in sports video, the current research on video detail enhancement technology does not pay much attention to the development of ball game violation tracking and recognition. Therefore, the study uses the fusion algorithm of wavelet exchange method and three-frame difference method and background subtraction method to detect and extract the motion targets, and uses the improved CamShift tracking algorithm and HMM to track and identify the tracking targets for the violation actions. Comprehensively, the study constructs a tracking recognition model for volleyball violation based on video enhancement technology to achieve accurate penalty in intense rivalry games. Through experimental analysis and comparison, the tracking F-measure value of the model constructed by the study is 0.89, which can achieve a good tracking effect, the recognition accuracy is 99.76%, and the average error is 0.003, which can effectively realize the tracking recognition of players’ illegal actions during volleyball, and objectively make court penalties to guarantee the fairness and justice of the game.

Author 1: Jieli Huang

Author 2: Wenjun Zou

Keywords: Volleyball; video detail enhancement; hmm; CamShift tracking; detection; recognition

Paper 71: Deep Learning-based Multiple Bleeding Detection in Wireless Capsule Endoscopy

Abstract: Wireless Capsule Endoscopy (WCE) is a diagnostic technology for gastrointestinal tract pathology detection. It has emerged as an alternative to conventional endoscopy which could be distressing to the patient. However, the diagnosis process requires to view and analyze hundreds of frames extracted from WCE video. This makes the diagnosis tedious. For this purpose, researches related to the automatic detection of signs of gastrointestinal diseases have been boosted. In this paper, we design a pattern recognition system for detecting Multiple Bleeding Spots (MBS) using WCE video. The proposed system relies on the Deep Learning approach to accurately recognize multiple bleeding spots in the gastrointestinal tract. Specifically, the You Only Look Once (YOLO) Deep Learning models are explored in this paper, namely, YOLOv3, YOLOv4, YOLOv5 and YOLOv7. The results of experiments showed that YOLOv7 is the most appropriate model for designing the proposed MBS detection system. Specifically, the proposed system achieved a mAP of 0.86, and an IoU of 0.8. Moreover, the results of the detection were enhanced by augmenting the training data to reach a mAP of 0.883.

Author 1: Ouiem Bchir

Author 2: Ghaida Ali Alkhudhair

Author 3: Lena Saleh Alotaibi

Author 4: Noura Abdulhakeem Almhizea

Author 5: Sara Mohammed Almuhanna

Author 6: Shouq Fahad Alzeer

Keywords: Wireless Capsule Endoscopy (WCE); Multiple Bleeding Spots (MBS); Gastrointestinal (GI) disease; deep learning; pattern recognition

Paper 72: A Novel Feature Fusion for the Classification of Histopathological Carcinoma Images

Abstract: Breast cancer is a significant global health concern, demanding advanced diagnostic approaches. Although traditional imaging and manual examinations are common, the potential of artificial intelligence (AI) and machine learning (ML) in breast cancer detection remains underexplored. This study proposes a hybrid approach combining image processing and ML methods to address breast cancer diagnosis challenges. The method utilizes feature fusion with gray-level co-occurrence matrix (GLCM), local binary patterns (LBP), and histogram features, alongside an ensemble learning technique for improved classification. Results demonstrate the approach's effectiveness in accurately classifying three carcinoma classes (ductal, lobular, and papillary). The Voting Classifier, an ensemble learning model, achieves the highest accuracy, precision, recall, and F1-scores across carcinoma classes. By harnessing feature extraction and ensemble learning, the proposed approach offers advantages such as early detection, improved accuracy, personalized medicine recommendations, and efficient analysis. Integration of AI and ML in breast cancer diagnosis shows promise for enhancing accuracy, effectiveness, and personalized patient care, supporting informed decision-making by healthcare professionals. Future research and technological advancements can refine AI-ML algorithms, contributing to earlier detection, better treatment outcomes, and higher survival rates for breast cancer patients. Validation and scalability studies are needed to confirm the effectiveness of the proposed hybrid approach. In conclusion, leveraging AI and ML techniques has the potential to revolutionize breast cancer diagnosis, leading to more accurate and personalized detection and treatment. Technology-driven advances can significantly impact breast cancer care and management.

Author 1: Salini S Nair

Author 2: M. Subaji

Keywords: Breast cancer; machine learning; artificial intelligence; feature extraction; ensemble classifier

Paper 73: Deep Conv-LSTM Network for Arrhythmia Detection using ECG Data

Abstract: In the evolving realm of medical diagnostics, electrocardiogram (ECG) data stands as a cornerstone for cardiac health assessment. This research introduces a novel approach, leveraging the capabilities of a Deep Convolutional Long Short-Term Memory (Conv-LSTM) network for the early and accurate detection of arrhythmias using ECG data. Traditionally, cardiac anomalies have been diagnosed through heuristic means, often requiring intricate scrutiny and expertise. However, the Deep Conv-LSTM model proposed herein addresses the inherent limitations of traditional methods by amalgamating the spatial feature extraction capability of convolutional neural networks (CNN) with the temporal sequence learning capacity of LSTM networks. Initial results derived from a diverse dataset, comprising myriad ECG waveform anomalies, delineated an enhancement in accuracy, reducing false positives and facilitating timely interventions. Notably, the model showcased adaptability in handling the burstiness of ECG signals, reflecting various heart rhythms, and the perplexity inherent in diagnosing subtle arrhythmic events. Additionally, the model's ability to discern longer, more complex patterns alongside transient anomalies offers potential for broader applications in telemetry and continuous patient monitoring systems. It is anticipated that this innovative fusion of CNN and LSTM architectures will usher a paradigm shift in automated arrhythmia detection, bridging the chasm between technology and the intricate nuances of cardiac physiology, thus improving patient outcomes.

Author 1: Alisher Mukhametkaly

Author 2: Zeinel Momynkulov

Author 3: Nurgul Kurmanbekkyzy

Author 4: Batyrkhan Omarov

Keywords: Deep learning; Conv-LSTM; classification; ECG; CNN

Paper 74: DetBERT: Enhancing Detection of Policy Violations for Voice Assistant Applications using BERT

Abstract: Voice Assistants, also known as VAs, have gained popularity in the last few years. They make our daily tasks easier via simple voice instructions. VAs platforms allow third-party developers to develop voice applications and publish them on the VAs platforms. However, VAs applications may collect users’ personal information for different purposes. To maintain the security and privacy of users, VAs platforms have specified a set of policies that must be adhered to by VAs applications’ developers. This paper aims to automatically detect voice apps that do not comply with the VA's platforms policies. To this end, DetBERT, a comprehensive testing tool, was built. DetBERT evaluates voice apps' compliance with the policies using BERT model by analyzing the apps’ behaviors and detecting violations. With DetBERT, a total of 50,000 voice assistant apps from Amazon Alexa and Google Assistant platforms were tested. The paper demonstrates that DetBERT can accurately identify whether a voice assistant application has violated the platform’s policy or not.

Author 1: Rawan Baalous

Author 2: Joud Alzahrani

Author 3: Mariam Ali

Author 4: Rana Asiri

Author 5: Eman Nooli

Keywords: Alexa; Google assistant; BERT; policy violation detector; voice assistant; user privacy; security

Paper 75: Digital Stethoscope for Early Detection of Heart Disease on Phonocardiography Data

Abstract: The burgeoning realm of digital healthcare has unveiled a novel diagnostic instrument: a digital stethoscope tailored for the early detection of heart disease as elucidated in this research. By harnessing the nuanced capabilities of phonocardiography, this device captures intricate heart sounds, subsequently processed through advanced machine learning algorithms. Traditional stethoscopes, although indispensable, might miss subtle anomalies – a lacuna this digital counterpart addresses by meticulously analyzing phonocardiographic data for the slightest deviations indicative of cardiac anomalies. As the digital stethoscope delves into this trove of aural cues, the machine learning component discerns patterns and irregularities often imperceptible to human auditors. The confluence of these digital acoustics and computational analytics not only augments the accuracy of early heart disease diagnosis but also facilitates the archival of this data, engendering a continuous, longitudinal assessment of cardiac health. The initial foray into real-world application registered an encouraging precision rate, cementing its potential as an invaluable asset in preemptive cardiac care. With this innovation, we stand on the cusp of a paradigm shift in how heart diseases are diagnosed, making strides towards timely interventions and improved patient outcomes.

Author 1: Batyrkhan Omarov

Author 2: Assyl Tuimebayev

Author 3: Rustam Abdrakhmanov

Author 4: Bakytgul Yeskarayeva

Author 5: Daniyar Sultan

Author 6: Kanat Aidarov

Keywords: Deep learning; CNN; random forest; SVM; neural network; prediction; analysis

Paper 76: Predicting the Level of Safety Feeling of Bangladeshi Internet users using Data Mining and Machine Learning

Abstract: An amazing combination of cutting-edge data mining and machine learning methodologies to predict the level of safety feeling among Bangladeshi internet users, which is a significant departure in this subject. By leveraging cutting-edge algorithms and innovative data sources, this work provides previously unheard-of insights into how this demographic perceives online safety, shedding light on an essential yet underappreciated aspect of their digital lives. This exceptional study's original research increases the body of knowledge of online safety and sets the road for policy recommendations and intervention tactics that will enable Bangladesh to become a global leader in internet security.

Author 1: Md. Safiul Alam

Author 2: Anirban Roy

Author 3: Partha Protim Majumder

Author 4: Sharun Akter Khushbu

Keywords: Bangladesh; data analysis; data mining; important factors; machine learning; prediction; performance evaluation metrics; safety level

Paper 77: A Novel Deep Neural Network to Analyze and Monitoring the Physical Training Relation to Sports Activities

Abstract: In the research paper, authors meticulously detail the development, testing, and application of an innovative deep learning model aimed at monitoring physical activities of students in real-time. Drawing upon the advanced capabilities of convolutional neural networks (CNNs), the proposed system exhibits an exceptional ability to track, analyze, and evaluate the physical exercises performed by students, thereby providing an unprecedented scope for customization in physical education strategies. This piece of scholarly work bridges the gap between physical education and cutting-edge technology, highlighting the burgeoning role of artificial intelligence in health and fitness sector. With an expansive study spanning various cohorts of physical culture students, the paper provides compelling empirical evidence that underlines the superiority of the deep learning system over conventional methods in aspects of accuracy, speed, and efficiency of monitoring. The authors demonstrate the transformative potential of their system, capable of facilitating personalized and optimized physical training strategies based on real-time feedback. Moreover, the potential implications of the study extend beyond the realm of education and into wider public health applications, with the possibility of fostering improved health outcomes on a larger scale. This research paper makes a significant contribution to the burgeoning field of AI in physical education, embodying a paradigm shift in the approach towards physical fitness and health monitoring. It underscores the potential of AI-driven technology to revolutionize traditional methods in physical education, paving the way for more personalized and effective teaching and training regimes, and ultimately contributing to enhanced health and fitness outcomes among students.

Author 1: Bakhytzhan Omarov

Author 2: Nurlan Nurmash

Author 3: Bauyrzhan Doskarayev

Author 4: Nagashbek Zhilisbaev

Author 5: Maxat Dairabayev

Author 6: Shamurat Orazov

Author 7: Nurlan Omarov

Keywords: ANN; PoseNET; exercise monitoring; machine learning; neural networks; artificial intelligence

Paper 78: Hybrid CNN-LSTM Network for Cyberbullying Detection on Social Networks using Textual Contents

Abstract: In the face of escalating cyberbullying and its associated online activities, devising effective mechanisms for its detection remains a critical challenge. This study proposes an innovative approach, integrating Long Short-Term Memory (LSTM) networks with Convolutional Neural Networks (CNN), for the detection of cyberbullying in online textual content. The method uses LSTM to understand the temporal aspects and sequential dependencies of text, while CNN is employed to automatically and adaptively learn spatial hierarchies of features. We introduce a hybrid LSTM-CNN model which has been designed to optimize the detection of potential cyberbullying signals within large quantities of online text, through the application of advanced natural language processing (NLP) techniques. The paper reports the results from rigorous testing of this model across an extensive dataset drawn from multiple online platforms, indicative of the current digital landscape. Comparisons were made with prevailing methods for cyberbullying detection, demonstrating a substantial improvement in accuracy, precision, recall and F1-score. This research constitutes a significant step forward in developing robust tools for detecting online cyberbullying, thereby enabling proactive interventions and informed policy development. The effectiveness of the LSTM-CNN hybrid model underscores the transformative potential of leveraging artificial intelligence for social safety and cohesion in an increasingly digitized society. The potential applications and limitations of this model, alongside avenues for future research, are discussed.

Author 1: Daniyar Sultan

Author 2: Mateus Mendes

Author 3: Aray Kassenkhan

Author 4: Olzhas Akylbekov

Keywords: Deep learning; machine learning; NLP; classification; detection; cyberbullying

Paper 79: Real-Time Road Surface Damage Detection Framework based on Mask R-CNN Model

Abstract: In the ever-evolving realm of infrastructure management, the timely and accurate detection of road surface damages is imperative for the longevity and safety of transportation networks. This research paper introduces a pioneering framework centered on the Mask R-CNN (Region-based Convolutional Neural Networks) model for real-time road surface damage detection. The overarching methodology encapsulates a deep learning-based approach to discern and classify various road aberrations such as potholes, cracks, and rutting. The chosen Mask R-CNN architecture, renowned for its proficiency in instance segmentation tasks, has been fine-tuned and optimized specifically for the unique challenges posed by road surfaces under diverse lighting and environmental conditions. A diverse dataset, amalgamating urban, suburban, and rural roadways under varied climatic conditions, served as the foundation for model training and validation. Preliminary results have not only underscored the model's robustness in real-time detection but also its superiority in terms of accuracy and computational efficiency when juxtaposed with extant methods. Concomitantly, the framework emphasizes scalability and adaptability, positing it as a frontrunner for potential integration into automated road maintenance systems and vehicular navigation aids. This trailblazing endeavor elucidates the potentialities of deep learning paradigms in revolutionizing road management systems, thus fostering safer and more efficient transportation environments.

Author 1: Bakhytzhan Kulambayev

Author 2: Magzat Nurlybek

Author 3: Gulnar Astaubayeva

Author 4: Gulnara Tleuberdiyeva

Author 5: Serik Zholdasbayev

Author 6: Abdimukhan Tolep

Keywords: Deep learning; CNN; random forest; SVM; neural network; prediction; analysis

Paper 80: Osteoporosis Detection and Classification of Femur X-ray Images Through Spectral Domain Analysis using Texture Features

Abstract: Osteoporosis commonly diagnosed as a bone disorder that affects the significant portion of the population. The Dual X-ray Absorptiometry (DXA) is one of the most accepted standard methods of analyzing the bone disorder, but it is exorbitant. However X-ray is a cost effective, therefore the proposed work introduces a new technique to improve osteoporosis detection and classification of femur bone X-ray image. The spectral based sub band images texture features are used to analyze the Region Of Interest (ROI) femoral head trabecular bone. A spectral domain based on the Two-Dimensional Discrete Wavelet Transform (2D-DWT) is used to represent variations in finer details in the image. Trabecular femur bone texture is determined only by horizontal, vertical, and diagonal sub bands of DWT coefficients. The sub band images are further enhanced by applying the maximum response filter (MRF) at different scales, thereby enhancing the most significant responses. Consequently, the sum of the MRFs of different scale images is considered as the supervised database. To detect osteoporosis, the test and supervised images are analyzed to calculate two significant attributes such as Zero Mean Normalized Cross-Correlation (ZMNC) and Sum Squared Difference (SSD). Based on experimental results, the performance metrics measure is improved in all aspects over current methods.

Author 1: Dhanyavathi A

Author 2: Veena M B

Keywords: Classification; feature; femur; images; normal; osteopenia; osteoporosis; texture

Paper 81: A Systematic Review on Blockchain Scalability

Abstract: Blockchain is an exciting new technology that has garnered attention across multiple industries. This new technology offers several advantages, including decentralization, transparency, and immutability. However, several issues limit the effectiveness of this technology, such as scalability, interoperability, and privacy. A systematic review of blockchain scalability research was conducted using three primary databases: ACM, Science Direct, and IEEE. The review examined the state of the art in blockchain scalability, identifying the most important research trends and challenges. The solutions that have been established can be categorized into two main groups: those that pertain to block storage and those that pertain to the underlying blockchain mechanism. Numerous solutions were suggested for each main group. The most common proposed solutions for improving the scalability of blockchain networks in the literature are improving the consensus algorithm and using sharding. Most of the solutions were proof of concept and need more investigation in the future.

Author 1: Asmaa Aldoubaee

Author 2: Noor Hafizah Hassan

Author 3: Fiza Abdul Rahim

Keywords: Blockchain; scalability; sharding; consensus algorithm

Paper 82: Machine Learning Techniques for Diabetes Classification: A Comparative Study

Abstract: In light of the growing global diabetes epidemic, there is a pressing need for enhanced diagnostic tools and methods. Enter machine learning, which, with its data-driven predictive capabilities, can serve as a powerful ally in the battle against this chronic condition. This research took advantage of the Pima Indians Diabetes Data Set, which captures diverse patient information, both diabetic and non-diabetic. Leveraging this dataset, we undertook a rigorous comparative assessment of six dominant machine learning algorithms, specifically: Support Vector Machine, Artificial Neural Networks, Decision Tree, Random Forest, Logistic Regression, and Naive Bayes. Aiming for precision, we introduced principal component analysis to the workflow, enabling strategic dimensionality reduction and thus spotlighting the most salient data features. Upon completion of our analysis, it became evident that the Random Forest algorithm stood out, achieving an exemplary accuracy rate of 98.6% when 'BP' and 'SKIN' attributes were set aside. This discovery prompts a crucial discussion: not all data attributes weigh equally in their predictive value, and a discerning approach to feature selection can significantly optimize outcomes. Concluding, this study underscores the potential and efficiency of machine learning in diabetes diagnosis. With Random Forest leading the pack in accuracy, there's a compelling case to further embed such computational techniques in healthcare diagnostics, ushering in an era of enhanced patient care.

Author 1: Hiri Mustafa

Author 2: Chrayah Mohamed

Author 3: Ourdani Nabil

Author 4: Aknin Noura

Keywords: Machine learning; support vector machine; artificial neural networks; decision tree; random forest; logistic regression; naive bayes; principal component analysis; classification; diabetes

Paper 83: An Optimized Survival Prediction Method for Kidney Transplant Recipients

Abstract: Human organ transplantation is a lifesaving process for many of the patients suffering from end stage diseases. Transplantation surgeons are often confronted with the question of the expected survival prognosis for this expensive and perilous process.The aim of the work is to identify an optimal model for predicting the survival of the recipient based on the available organ. This study identifies important features of the recipient and donor parameters for training the model. The study compares the performance of the Random Survival Forest (RSF), which is a machine learning method, and the Cox Proportional Hazard (CPH) model, which is a statistical model, to identify the more accurate model for survival prediction. Variations of the C-index, Brier score, and cumulative Area Under Curve evaluate the survival models considered. This study suggests that CPH which is a statistical method is a better option for forecasting graft and patient survival for an improved clinical outcome.

Author 1: Benita Jose Chalissery

Author 2: V. Asha

Keywords: Cox proportional hazard model; random survival forest; C-index; brier score; area under curve; organ transplantation; survival prognosis

Paper 84: Analyzing RNA-Seq Gene Expression Data for Cancer Classification Through ML Approach

Abstract: Purpose: Ribonucleic Acid Sequencing (RNA-Seq) is a technique that allows an efficient genome-wide analysis of gene expressions. Such analysis is a strategy for identifying hidden patterns in data, and those related to cancer-specific biomarkers. Prior analyses without samples of different cancer kinds used RNA-Seq data from the same type of cancer as the positive and negative samples. Therefore, different cancer types must be evaluated to uncover differentially expressed genes and perform multiple cancer classifications. Problem: Since gene expression reflects both the genetic make-up of an organism and the biochemical activities occurring in tissue and cells, it can be crucial in the early identification of cancer. The aim of this study is to classify the RNA-Sequence data into five different cancer forms, such as LUAD, BRCA, KIRC, LUSC, and UCEC, through an ensemble approach of machine learning algorithms. RNA-Seq data for five different cancer types from the UCI Machine Learning Repository are examined in this research. Methods: As a first step, the relevant features of RNA-Seq are extricated using Principal Component Analysis (PCA). Then, the extricated features are given to the ensemble of machine learning classifiers to classify the type of cancer. The ensemble of classifiers is built using Support Vector Machine (SVM), Naive Bayes (NB), and K-Nearest Neighbor (KNN). Results: The results demonstrated that the proposed ensemble classifier outperformed the existing machine-learning approaches with an accuracy of 99.59%.

Author 1: Abdul Wahid

Author 2: M Tariq Banday

Keywords: RNA-Sequence; gene expression; feature extraction; voting classifier; ensemble approach

Paper 85: Historical Building 3D Reconstruction for a Virtual Reality-based Documentation

Abstract: An innovative preservation approach was proposed to document historical buildings in 3D model, and to present it virtually. The approach was applied to the Lawang Sewu building, one of the architectural masterpieces that is part of Indonesian history. Virtual Reality (VR) technology was used to create a Lawang Sewu VR application program that allows users to virtually walk around the building. A new method for 3D reconstruction was proposed, where data of photo, video and miniature documentation, as well as notes collected from observations were used as the main reference. Meanwhile, architectural record data was used in cases where information cannot be obtained through the main reference. The proposed method focuses on traditional techniques, both at the data acquisition and 3D modelling stages. Poly modelling techniques were chosen for 3D reconstruction. The poly modelling technique was chosen based on its ease and flexibility in controlling the number of polys in 3D models, and was suitable to be applied for repetitive spatial typologies, such as the Lawang Sewu building. After given textures, the 3D model was sent to the VR editor. In addition of running on the desktop platform, Head Mounted Device (HMD) that supports the creation of an immersive experience, was also chosen to run the Lawang Sewu VR. The evaluation carried out to measure the level of similarity of the 3D model to the original building and the sensation of an immersive experience felt by the user shows good achievements.

Author 1: Ahmad Zainul Fanani

Author 2: Arry Maulana Syarif

Keywords: Virtual reality; immersive presentation; 3D reconstruction; historical heritage building preservation

Paper 86: Identifying and Prioritizing Digital Transformation Elements Using Fuzzy Analytic Hierarchy Process

Abstract: Digital transformation addresses multiple aspects of the organization. These aspects are the elements to be addressed for the digital transformation in any organization and are categorized as dimensions and sub-dimensions. In this work, these elements are collected from a wide range of related literature (56 publications). The most relevant elements were then identified through expert survey; involving 12 experts. The weights for these elements were identified using multi-criteria decision-making (MCDM) techniques. The Analytical Hierarchy Process (AHP) is one of the most often used MCDM techniques to incorporate individual and subjective preferences when conducting analysis and convert complex issues into a clear hierarchical structure. This work applies fuzzy AHP to take into consideration the treatment of uncertainty issues (in AHP), using the geometric mean method, and through an iterative process, calculate the weights of various dimensions and sub-dimensions, and prioritize them within the proposed roadmap for digital transformation implementation. Sensitivity analysis and comparison with AHP were used to validate our findings and the robustness of our approach. The proposed approach identified 9 main dimensions and 42 sub-dimensions which align with the majority of the literature. However, the advantage of this approach is the prioritization of these nine dimensions and their sub-dimensions as per the weights assigned to each one of them, allowing the project manager to allocate the available resources to the dimensions with the highest priority. The results show that the strategy and business process dimensions are the most crucial ones in the implementation of digital transformation.

Author 1: Mohammed Hitham M. H

Author 2: Hatem Elkadi

Author 3: Neamat El Tazi

Keywords: Digital transformation; MCDM; AHP; fuzzy AHP introduction

Paper 87: Machine Learning based Predictive Modelling of Cybersecurity Threats Utilising Behavioural Data

Abstract: With the rapid advancement of technology in Malaysia, the number of cybercrimes is also increasing. To stop the increase in cybercrimes, everyone, including normal citizens, needs to know how secure they are while using digital appliances. A system is developed to predict the risk of users based on their behaviour when they are online using real-life behavioural data obtained from a private university’s 207 undergraduates. Five supervised machine learning methods are being tested which are: Regression Logistics, K-Nearest Neighbour (KNN), Decision Tree (DT), Support Vector Machine (SVM), and Naïve Bayesian Classifier with the aid of a tool, RapidMiner. The algorithms are used to construct, test, and validate three categories of cybercrime threat (Malware, Social Engineering, and Password Attack) predictive models. It was found that KNN model produces the highest accuracy and lowest classification error for all three categories of cybercrime threat. This system is believed to be crucial in alerting users with details of whether the consumer behaviour risk is high or low and what further actions can be taken to increase awareness. This system aims to prevent the rise in cybercrimes by providing a prediction of their risk levels in cybersecurity to encourage them to be more proactive in cybersecurity.

Author 1: Ting Tin Tin

Author 2: Khiew Jie Xin

Author 3: Ali Aitizaz

Author 4: Lee Kuok Tiung

Author 5: Teoh Chong Keat

Author 6: Hasan Sarwar

Keywords: Cybersecurity threat; cybersecurity risk; predictive modeling; undergraduates; cybercrime

Paper 88: Enterprise Marketing Decision: Advertising Click Through Rate Prediction Based on Deep Neural Networks

Abstract: With the high-speed growth of modern information technology, online advertising, as a new form of advertising on the Internet, has begun to emerge, demonstrating enormous development potential. To improve the accurate estimation of advertising placement and improve the operational efficiency of the advertising placement system, an improved deep neural network model for forecasting advertising click through rate was studied and designed. Meanwhile, the values of the activation function and the parameter dropout are determined, and the prediction accuracy of the deep neural network model and the improved model is compared and analyzed. The experimental results show that the training time of the improved prediction model has been shortened by about 73.25%, resulting in a significant improvement in computational efficiency. When the number of iterations is 110, the logarithmic loss function value is 0.208, and the logarithmic loss function value of the improved model is 0.207, with an average loss reduction of 0.4%. In the area comparison under the receiver operating characteristic curve, the pre improved model was 0.7092, and the improved model was 0.7207. Meanwhile, compared to before the improvement, the prediction accuracy of the improved model increased by 1.6%. The data validates that the optimized model has high prediction precision and efficiency, and has certain application potential and commercial value in marketing.

Author 1: Luyao Zhan

Keywords: Click through rate prediction; deep learning; deep neural network; online advertising; marketing

Paper 89: Design and Development of an Intelligent Rendering System for New Year's Paintings Color Based on B/S Architecture

Abstract: With the arrival of the synthetic talent era, laptop technological know-how for the safety and inheritance of intangible cultural heritage has added a new way of thinking, and the range of intangible cultural heritage additionally offers greater chances for laptop technology, the utility of laptop talent science to New Year's Eve artwork of the applicable lookup there are many gaps. Training of Cyclic Generative Adversarial Network (CycleGAN) realize the task of extracting plots of different site types from planar maps and the Rendering generation from planar color block map to color texture map. This paper first introduces the B/S community architecture, Python programming technological know-how and Django framework. Then the unique approach of using pc Genius to the project of rendering Chinese New Year artwork is clarified via modeling, studying algorithms, and community architecture. Finally, a hierarchical fusion generative adversarial neural community structure is designed primarily based on generative adversarial neural networks. The structural and textural features of the image are fused by texture GAN and then rendered to generate the New Year paintings. The test results show that this kind of algorithm draws clear texture, realistic images and full color of the New Year's pictures, and the IS index reaches 3.16 in the quantitative analysis, which is higher than other comparison algorithms.

Author 1: Zaozao Guo

Keywords: B/S architecture; intelligent rendering; adversarial neural network; Chinese New Year painting

Paper 90: Using EEG Effective Connectivity Based on Granger Causality and Directed Transfer Function for Emotion Recognition

Abstract: Emotion is a complex phenomenon that originates from everyday issues and has significant effects on individual decisions. Electroencephalography (EEG) is one of the widely used tools in examining the neural correlates of emotions. In this research, two concepts of Granger causality and directional transfer function were utilized to analyze EEG data recorded from 36 healthy volunteers in positive, negative and neutral emotional states and determine the effective connectivity between different brain sources (obtained through independent component analysis). Shannon entropy was utilized to sort the brain sources obtained by the ICA method, and average topography helps to add spatial information to the proposed connectivity models. According to the obtained confusion matrix, our method yielded an overall accuracy of 75% in recognizing three emotional states. Positive emotion was recognized with the highest accuracy of 87.96% (precision = 0.78, recall = 0.78 and F1-score = 0.81), followed by neutral (accuracy = 82.41%) and negative (accuracy = 79.63%) emotions. Indeed, our proposed method achieved the highest recognition accuracy for positive emotion. The proposed model in the present study has the ability to identify emotions in a completely personalized way based on neurobiological data. In the future, the proposed approach in the present study can be integrated with machine learning and neural network methods.

Author 1: Weisong Wang

Author 2: Wenjing Sun

Keywords: EEG; effective connectivity; granger causality; directed transfer function; emotion recognition

Paper 91: Development of an Image Encryption Algorithm using Latin Square Matrix and Logistics Map

Abstract: The goal of this study was to develop a robust image cryptographic scheme based on Latin Square Matrix and Logistics Map, capable of effectively securing sensitive data. Logistics mapping is a comparatively strong chaos system which enciphers with an unpredictability that significantly reduces the chance of deciphering. Additionally, the Latin square matrix stands out for its uniform histogram distribution, thereby bolstering its encryption's potency. The consequent integration of these algorithms in this study was therefore grounded in the scientific rationale of establishing a strong and resilient cypher technique. The study provides a new chaos-based method and extends the application of the probabilistic approach to the domain of symmetric key image encryption. Permutation and substitution approaches of image encryption were deployed to address the issue of images volume and differing sizes. The issue of misplaced pixel positions in the image was also adequately addressed, making it an effective method for image encryption. The hybrid technique was simulated on image data and evaluated to gauge its performance. Results showed that the algorithm was able to securely protect image data and the private information associated with them, while also making it very difficult for unauthorized users to decrypt the information. The average encryption time of 184(μs) on seven (7) images showed that it could to be deployed for real-time systems. The proposed method obtained an average entropy of 7.9398 with key space of 1.17x1077 and an average avalanche effect (%) of 49.9823 confirming the security and resilience of the developed method.

Author 1: Emmanuel Oluwatobi Asani

Author 2: Godsfavour Biety-Nwanju

Author 3: Abidemi Emmanuel Adeniyi

Author 4: Salil Bharany

Author 5: Ashraf Osman Ibrahim

Author 6: Anas W. Abulfaraj

Author 7: Wamda Nagmeldin

Keywords: Image encryption; algorithm; logistics map; Latin square matrix; chaos technology

Paper 92: Tampering Detection and Segmentation Model for Multimedia Forensic

Abstract: When an image undergoes hybrid post-processing transformation, detecting tamper region, localizing it and segmentation becomes very difficult tasks. In particular, when a copy-move attack with hybrid transformation has similar contrast and illumination parameters with an authenticated image it makes tamper detection difficult. Alongside, under small-smooth attack existing tamper identification model provides a very poor segmentation outcome and sometimes fails to identify an image as tampered. This article focused on addressing the difficulty through the adoption of the Deep Learning model. The proposed technique is efficient in detecting tampering with good segmentation outcomes. However, existing models fail to distinguish adjacent pixels' relationships affecting segmentation outcomes. In this paper, an Improved Convolution Neural Network (ICNN) assuring correlation awareness-based Tamper Detection and Segmentation (TDS) model for image forensics is presented. This model brings good correlation among adjacent pixels through the introduction of an additional layer namely the correlation layer alongside vertical and horizontal layers. The TDS-ICNN is very effective in localizing and segmenting tamper regions even under small-smooth post-processing tampering attacks by using a feature descriptor built using aggregated three-layer ICNN architecture. An experiment is done to study TDS-ICNN with other tamper identification models using various datasets such as MICC, Coverage, and CoMoFoD. The TDS-ICNN is very efficient under different post-processing hybrid attacks when compared with existing models.

Author 1: Manjunatha S

Author 2: Malini M Patil

Author 3: Swetha M D

Author 4: Prabhu Vijay S S

Keywords: Convolution neural networks; digital image forensic; hybrid image transformation; resampling feature; segmentation

Paper 93: PRESSNet: Assessment of Building Damage Caused by the Earthquake

Abstract: Loss of life and property often occur due to natural disasters and other significant occurrences like earthquakes, which make manual damage assessment a time-consuming and inefficient process. In an attempt to address this challenge, researchers have been investigating the field of automated damage assessment in Remote Sensing. With time, this area of research has transformed from conventional machine learning techniques to more sophisticated deep learning techniques. The study puts forward the PRESSNet model as a solution for assessing building damage. The effectiveness of the proposed PRESSNet model is compared to that of a baseline model, PSPNet, and ResNet 50, across different types of damage. This study contributes by introducing the spatial attention module to the baseline model. The xBD Dataset was used both before and after the Palu earthquake disaster. The results show that PRESSNet performs similarly or slightly better than the baseline model in all damage categories. This illustrates the impressive ability of the proposed PRESSNet architecture to accurately detect and classify building damage. This research sheds light on the development of effective models for assessing disaster damage and lays the foundation for future progress in this crucial area.

Author 1: Dewa Ayu Defina Audrey Nathania

Author 2: Alexander Agung Santoso Gunawan

Author 3: Edy Irwansyah

Keywords: Remote sensing; deep learning; PSPNet; ResNet; spatial attention

Paper 94: Group Intelligence Recommendation System based on Knowledge Graph and Fusion Recommendation Model

Abstract: The challenge of how to further improve the accuracy of the system's recommendations in a data-limited environment is crucial as the use of group intelligence recommendation systems in everyday life increases. Through the fusion of different types of auxiliary information, this study develops a multi-feature fusion model based on the conventional recommendation model by introducing knowledge graphs. It also considers the homogeneity of push results caused by graph convolutional network smoothing when using knowledge graphs, and designs a fusion label propagation algorithm and graph convolution. The multi-feature fusion model had a maximum hit rate of over 80% and a normalised discount gain of up to 43% running time much lower than the conventional graph convolution recommendation model in the representation dimension interval [2, 32], while the fusion label propagation algorithm and graph convolution network model maintained a hit rate and normalised discount gain higher than the conventional model by 2 to 1 under 10 consecutive epochs. With a hit rate and normalised discount gain 2 to 10 percentage points higher than the conventional model, the coverage rate increased to 49.8%. This study is useful for research on group intelligence recommendation systems and can serve as a technical guide for improving the ability of group intelligence systems to make recommendations quickly.

Author 1: Chengning Huang

Author 2: Bo Jing

Author 3: Lili Jiang

Author 4: Yuquan Zhu

Keywords: Knowledge graphs; recommendation system; graph convolutional networks; label propagation algorithms

Paper 95: Statistical Language Model-based Analysis of English Corpora and Literature

Abstract: Despite widespread use of statistical language models in language processing, their ability to process natural languages is not advanced and they struggle to effectively capture linguistic information. Furthermore, there is a lack of automatic processing models in the field of natural language processing. In order to address these issues, and Improve the processing ability of statistical language models for English language a statistical language model optimization algorithm has been proposed. This algorithm is based on an improved resorting algorithm and is specifically applied to process English literary texts. Experimental results indicate that the proposed algorithm outperforms the N-gram algorithm in a majority of texts, with a maximum accuracy improvement of 14.5%. Additionally, in terms of the grammar analysis model, there is a high level of consistency between the model's scoring and the expert manpower scoring, as reflected by a correlation coefficient of 0.7893. This high level of consistency between the grammar analysis model and expert analysis results holds significant importance for the advancement of natural language processing.

Author 1: Wenwen Chai

Keywords: Statistical language model; corpus; English literature; reordering; grammatical analysis

Paper 96: Marginal Distribution Algorithm for Feature Model Test Configuration Generation

Abstract: Generating test configuration for Software Product Line (SPL) is difficult, due to the exponential effect of feature combination. Pairwise testing can generate test input for a single software product that deviates from exhaustive testing, nevertheless proven to be effective. In the context of SPL testing, to generate minimal test configuration that maximizes pairwise coverage is not trivial, especially when dealing with a huge number of features and when constraints must be satisfied, which is the case in most SPL systems. In this paper, we propose an estimation of distribution algorithm, based on pairwise testing, to alleviate this problem. Comparisons are made against a greedy-based and a constraint handling based approach. The experiments demonstrate the feasibility of the proposed algorithm, such that it achieves better test configurations dissimilarity and at the same time maintain the test configuration size and pairwise coverage. This is supported by analysis using descriptive statistics.

Author 1: Mohd Zanes Sahid

Author 2: Mohd Zainuri Saringat

Author 3: Mohd Hamdi Irwan Hamzah

Author 4: Nurezayana Zainal

Keywords: Estimation of distribution algorithm; marginal distribution algorithm; test configuration generation; pairwise testing; software product line

Paper 97: A QoS-Aware Resource Allocation Method for Internet of Things using Ant Colony Optimization Algorithm and Tabu Search

Abstract: In today's computing era, the Internet of Things (IoT) stands out for its implementation of automation, high-quality ecosystems, creative and efficient services, and higher productivity. IoT has found applications in various fields, such as education, healthcare, agriculture, military, and industry, where diverse resource requirements present a major challenge. To address this issue, we propose a novel QoS-aware resource allocation method for IoT systems. Our approach combines the Ant Colony Optimization (ACO) and Tabu Search (TS) algorithms to manage resources effectively, minimize energy consumption, reduce communication delays, and enhance overall system performance. Experimental results demonstrate the efficiency and effectiveness of our approach, with significant improvements in QoS metrics compared to traditional methods. By merging ACO and TS algorithms, our research contributes to the advancement of IoT capabilities, energy conservation, and business optimization.

Author 1: Shuling YIN

Author 2: Renping YU

Keywords: Internet of things; resource allocation; virtualization; Ant Colony Optimization; Tabu Search

Paper 98: Artificial Rabbits Optimizer with Deep Learning Model for Blockchain-Assisted Secure Smart Healthcare System

Abstract: Smart healthcare is based on the electronic health and medical histories of residents, combined with information technology (IT) which can be used to construct a variety of systems including humanised health management systems and convenient medical service systems. The transparency, traceability, decentralization and security of BC technology and machine learning (ML) will enable the medical sector to upgrade and optimise different forms of quality and service. Therefore, this study introduces an artificial rabbit optimizer with deep learning for Blockchain Assisted Secure Smart Healthcare System (ARODL-BSSHS) technique. The presented ARODL-BSSHS technique designs a new healthcare monitoring technique by using blockchain (BC) technology and classifies the presence of malicious activities in the healthcare system, and takes needed actions to predict the disease. For intrusion detection, the ARODL-BSSHS technique exploits the ARO algorithm with Hop field neural network (IHNN) model. On the other hand, the ARODL-BSSHS technique applies a deep extreme learning machine (DELM) model for disease detection purposes. Finally, the heap-based optimization (HBO) technique is exploited as a hyperparameter optimizer for the DELM model. The ARODL-BSSHS technique involves BC technology for the secure transmission of healthcare data. A series of simulations were carried out on benchmark datasets: heart disease and NSL-KDD database for examining the performance of the ARODL-BSSHS technique. The experimental values highlighted that the ARODL-BSSHS method obtains superior performance than other approaches.

Author 1: Mousa Mohammed Khubrani

Keywords: Blockchain; smart healthcare; artificial rabbit’s optimizer; deep learning; intrusion detection

Paper 99: A QoS-aware Mechanism for Reducing TCP Retransmission Timeouts using Network Tomography

Abstract: A wide range of web-based applications uses the Transmission Control Protocol (TCP) to ensure network resources are shared efficiently and fairly. As wired and wireless networks have become more complex, various end-to-end Congestion Control (CC) schemes have been developed, offering solutions through their proposed TCP variants. Network tomography, a powerful analytical tool, offers a unique perspective by measuring end-to-end performance to estimate internal network parameters, including latency. This estimation capability proves valuable, especially in cases where precise protocol performance evaluation is essential. TCP protocol can be improved significantly by properly estimating RTT time. It has resulted in better network conditions and improved reliability, as well as a higher level of user satisfaction. In this study, we propose a method to infer the link delay using network tomography and then adjust the RTT based on the delay estimation obtained in the previous step. Simulation results performed using the NS2 software show that the proposed method significantly improves the TCP protocol's Round-Trip Time (RTT) estimation by more than 15%. It reduces congestion, improves information transfer efficiency, and ensures the highest level of service in the network.

Author 1: Jingfu LI

Keywords: Latency; network tomography; end to end; depending on the probe

Paper 100: Routing Strategies and Protocols for Efficient Data Transmission in the Internet of Vehicles: A Comprehensive Review

Abstract: The Internet of Vehicles (IoV) integrates wireless communication, vehicular technology, and the Internet to create intelligent transportation systems. Efficient routing of data packets within the IoV is crucial for seamless communication and service enablement. This paper provides a comprehensive review of routing strategies and protocols in the IoV environment, categorizing and evaluating existing approaches. Routing protocols are classified, their adaptability is assessed to network variations, and their performance is compared. Insights are drawn from researchers' experiences. The paper offers a taxonomy of routing protocols, highlights adaptability to network conditions, and presents a comparative analysis. Lessons from researchers shed light on practical implications. The review identifies key routing challenges in IoV and provides a valuable resource for understanding and addressing these challenges in future research.

Author 1: Yijun Xu

Keywords: Internet of things; internet of vehicles; Vehicular Ad Hoc Networks (VANETs); routing; network adaptability; vehicular technology

Paper 101: Design and Implementation Submarine Cable Object Detection YOLOv4 based with Graphical User Interface (GUI) for Remotely Operated Vehicle (ROV)

Abstract: The use of submarine cables as underwater transmission channels for distributing electrical energy in Indonesian waters is crucial. However, the detection and maintenance of submarine cables still heavily rely on human observation, leading to limitations in time and subjective interpretations. This research aims to design and implement an underwater object detection system based on YOLOv4 integrated with a Graphical User Interface (GUI) on a Remotely Operated Vehicle (ROV) for submarine cable detection. The YOLOv4 model was trained using a balanced dataset, achieving performance with precision of 0.89, recall of 0.85, and f1-score of 0.87. Detection of Good Condition (SC-Good-Condition) achieved an Average Precision (AP) of 97.62%, while Bad Condition detection (SC-Bad-Condition) had an AP of 87.54%, resulting in an overall mAP of 92.58%. The implemented GUI successfully detected submarine cables in two test videos with FPS rates of 0.178 and 0.083. The designed underwater object detection system using YOLOv4 and GUI on ROV demonstrated satisfactory performance in detecting submarine cables. However, further efforts are needed to improve the GUI's FPS to make it more responsive and efficient. This research contributes to the development of underwater detection technology that supports environmental observation and electrical energy distribution in Indonesian waters.

Author 1: Fikri Arif Wicaksana

Author 2: Eueung Mulyana

Author 3: Syarif Hidayat

Author 4: Rahadian Yusuf

Keywords: Submarine cable; object detection; GUI; ROV; YOLOv4

Paper 102: Research on Clothing Color Classification Method based on Improved FCM Clustering Algorithm

Abstract: In the apparel industry, apparel color is an important factor to enhance the market competitiveness of enterprise products. However, the current prediction samples of clothing fashion color styling information do not incorporate practical cutting-edge fashion information. Therefore, Self-adaptive Weighted Kernel Function (SWK) has been introduced to traditional Fuzzy C-Means (FCM) clustering algorithms. After improvement, the SWK-FCM clustering algorithm is obtained, which enhances the classification ability of fashion colors and hue. Two prediction models have been developed using the finalized data of the International Fashion Color Committee, along with the SWK-FCM clustering algorithm. The models have been tested via experiments to verify their accuracy. The experimental results show that the classification coefficients of SWK-FCM clustering algorithm are 0.9553 and 0.9258 under 5% Gaussian noise. They are higher than those of FCM (0.7063) and FLICM (0.8598). The classification entropy is lower than that of the comparison algorithm, while the same results are presented under other conditions and in the actual experiments. In addition, the overall MSE of the GM (1, 1) prediction model using the final case information is 0.00028, which is close to the order of 10-4. The MSE value of the BP neural network prediction model using the final case information ranges from 0.000529 to 0.011025. Overall, the clustering algorithm of SWK-FCM has good classification performance. Additionally, the GM (1,1) model based on SWK-FCM has better prediction results, which can be effectively applied in practical clothing color classification and popular color prediction.

Author 1: Jinliang Liu

Keywords: Fuzzy clustering; SWK-FCM; fashion color scheme; Gaussian noise

Paper 103: Next-Generation Intrusion Detection and Prevention System Performance in Distributed Big Data Network Security Architectures

Abstract: Big data systems are expanding to support the rapidly growing needs of massive scale data analytics. To safeguard user data, the design and placement of cybersecurity systems is also evolving as organizations to increase their big data portfolios. One of several challenges presented by these changes is benchmarking real-time big data systems that use different network security architectures. This work introduces an eight-step benchmark process to evaluate big data systems in varying architectural environments. The benchmark is tested on real-time big data systems running in perimeter-based and perimeter-less network environments. Findings show that marginal I/O differences exist on distributed file systems between network architectures. However, during various types of cyber incidents such as distributed denial of service (DDoS) attacks, certain security architectures like zero trust require more system resources than perimeter-based architectures. Results illustrate the need to broaden research on optimal benchmarking and security approaches for massive scale distributed computing systems.

Author 1: Michael Hart

Author 2: Rushit Dave

Author 3: Eric Richardson

Keywords: Big data systems; zero trust architecture; benchmarking; distributed denial of service attacks

Paper 104: Machine Learning for Smart Cities: A Comprehensive Review of Applications and Opportunities

Abstract: The smart city concept originated a few years ago as a combination of ideas about how information and communication technologies can improve urban life. With the advent of the digital revolution, many cities globally are investing heavily in designing and implementing smart city solutions and projects. Machine Learning (ML) has evolved into a powerful tool within the smart city sector, enabling efficient resource management, improved infrastructure, and enhanced urban services. This paper discusses the diverse ML algorithms and their potential applications in smart cities, including Artificial Intelligence (AI) and Intelligent Transportation Systems (ITS). The key challenges, opportunities, and directions for adopting ML to make cities smarter and more sustainable are outlined.

Author 1: Xiaoning Dou

Author 2: Weijing Chen

Author 3: Lei Zhu

Author 4: Yingmei Bai

Author 5: Yan Li

Author 6: Xiaoxiao Wu

Keywords: Smart city; machine learning; artificial intelligence; intelligent transportation system; smart grids

Paper 105: Two Dimensional Deep CNN Model for Vision-based Fingerspelling Recognition System

Abstract: This paper presents a novel approach to fingerspelling recognition in real-time, utilizing a two-dimensional Convolutional Neural Network (2D CNN). Existing recognition systems often fall short in real-world conditions due to variations in illumination, background, and user-specific characteristics. Our method addresses these challenges, delivering significantly improved performance. Leveraging a robust 2D CNN architecture, the system processes image sequences representing the dynamic nature of fingerspelling. We focus on low-level spatial features and temporal patterns, thereby ensuring a more accurate capture of the intricate nuances of fingerspelling. Additionally, the incorporation of real-time video feed enhances the system's responsiveness. We validate our model through comprehensive experiments, showcasing its superior recognition rate over current methods. In scenarios involving varied lighting, different backgrounds, and distinct user behaviors, our system consistently outperforms. The findings demonstrate that the 2D CNN approach holds promise in improving fingerspelling recognition, thereby aiding communication for the hearing-impaired community. This work paves the way for further exploration of deep learning applications in real-time sign language interpretation. This research bears profound implications for accessibility and inclusivity in communication technology.

Author 1: Zhadra Kozhamkulova

Author 2: Elmira Nurlybaeva

Author 3: Leilya Kuntunova

Author 4: Shirin Amanzholova

Author 5: Marina Vorogushina

Author 6: Mukhit Maikotov

Author 7: Kaden Kenzhekhan

Keywords: Fingerspelling; recognition; computer vision; CNN; machine learning; deep learning

Paper 106: Non-contact Respiratory Rate Monitoring Based on the Principal Component Analysis

Abstract: Assessing respiratory rate is a critical determinant of one’s health status. The proposed approach relies on principal component analysis (PCA) for the continuous monitoring of breathing rate using an RGB camera. This method employs re-mote plethysmography, a video-based technique enabling contact-less tracking of blood volume fluctuations by detecting variations in pixel intensity on the skin. These pixels encompass the red, blue, and green channels, whose values, post-PCA dimensionality reduction, encode the signal containing vital information about the breathing rate. To assess the method’s performance, it was tested on a group of seven volunteers, including individuals of both genders. The results reveal a Mean Absolute Deviation of 0.714 BPM and a Root Mean Square Error of 2.035 BPM when comparing the experimental measurements to the actual readings.

Author 1: Hoda El Boussaki

Author 2: Rachid Latif

Author 3: Amine Saddik

Author 4: Zakaria El Khadiri

Author 5: Hicham El Boujaoui

Keywords: RGB; breathing rate; non-contact; principal component analysis; plethysmography

Paper 107: An Improved Genetic Algorithm with Chromosome Replacement and Rescheduling for Task Offloading

Abstract: End-Edge-Cloud Computing (EECC) has been applied in many fields, due to the increased popularity of smart devices. But the cooperation of end devices, edge and cloud resources is still challenge for improving service quality and resource efficiency in EECC. In this paper, we focus on the task offloading to address the challenge. We formulate the offloading problem as mixed integer nonlinear programming, and solve it by Genetic Algorithm (GA). In the GA-based offloading algorithm, each chromosome is the code of a offloading solution, and the evolution is to iteratively search the global best solution. To improve the performance of GA-based task offloading, we integrate two improvement schemes into the algorithm, which are the chromosome replacement and the task rescheduling, respectively. The chromosome replacement is to replace the chromosome of every individual by its better offspring after every crossing, which substitutes the selection operator for population evolution. The task rescheduling is rescheduling each rejected task to available resources, given offloading solution from every chromosome. Extensive experiments are conducted, and results show that our proposed algorithm can improve upto 32% user satisfaction, upto 12% resource efficiency, and upto 35.3% processing efficiency, compared with nine classical and up-to-date algorithms.

Author 1: Hui Fu

Author 2: Guangyuan Li

Author 3: Fang Han

Author 4: Bo Wang

Keywords: Genetic algorithm; task offloading; task scheduling; edge computing; cloud computing

Paper 108: An Efficient Convolutional Neural Network Classification Model for Several Sign Language Alphabets

Abstract: Although deaf people represent over 5% of the world’s population, according to what the World Health Organization stated in May 2022, they suffer from social and economic marginalization. One way to improve the lives of deaf people is to try to make communication between them and others easier. Sign language, the means through which deaf people can communicate with other people, can benefit from modern techniques in machine learning. In this study, several convolutional neural networks (CNN) models are designed to develop an efficient model, in terms of accuracy and computational time, for the classification of different signs. This research presents a methodology for developing an efficient CNN architecture from scratch to classify multiple sign language alphabets, which has numerous advantages over other contemporary CNN models in terms of prediction time and accuracy. This framework analyses the effect of varying CNN hyper-parameters, such as kernel size, number of layers, and number of filters in each layer, and picks the ideal parameters for CNN model construction. In addition, the suggested CNN architecture operates directly on unprocessed data without the need for preprocessing to generalize it across other datasets. In addition, the capacity of the model to generalize to diverse sign languages is rigorously evaluated using three distinct sign language alphabets and five datasets, namely, Arabic (ArSL), two American English (ASL), Korean (KSL), and the combination of Arabic and American datasets. The proposed CNN architecture (SL-CNN) outperforms state-of-the-art CNN models and traditional machine learning models achieving an accuracy of 100%, 98.47%, 100%, and 99.5% for English, Arabic, Korean, and combined Arabic-English alphabets, respectively. The prediction or inference time of the model is about three milliseconds on average, making it suitable for real-time applications. So, in the future, it is easy to turn this model into a mobile application.

Author 1: Ahmed Osman Mahmoud

Author 2: Ibrahim Ziedan

Author 3: Amr Ahmed Zamel

Keywords: Convolutional neural network (CNN); sign language; Arabic sign language (ArSL); American sign language (ASL); Korean sign language (KSL); Complexity time

Paper 109: Enhancing Outdoor Mobility and Environment Perception for Visually Impaired Individuals Through a Customized CNN-based System

Abstract: Visual impairment indicates any kind of vision loss including blindness. Individuals with visual impairments face significant challenges when trying to perceive their surroundings from a global perspective and navigating unfamiliar environments. Existing assistive technologies predominantly focus on obstacle avoidance, neglecting to provide comprehensive information about the overall environment. To address this gap, the proposed system employs a customized Convolutional Neural Network (CNN) model tailored to accurately predict the type of outdoor ground terrain the user is traversing. This information is then conveyed to the user audibly. It can also detect the presence of puddles on the road and let the user know whether the outside floor is wet (slippery). The proposed deep-learning architecture is trained on images collected from sources including the Stagnant Water dataset, the GTOS-Mobile dataset and a custom dataset. The trained model is then integrated into an Android app, providing visually impaired (VI) people with effective surrounding perception capabilities, leading to better travel and, ultimately, better living.

Author 1: Athulya N K

Author 2: Sivakumar Ramachandran

Author 3: Neetha George

Author 4: Ambily N

Author 5: Linu Shine

Keywords: Visually impaired; Terrain identification; Puddle detection; Deep learning

Paper 110: ArCyb: A Robust Machine-Learning Model for Arabic Cyberbullying Tweets in Saudi Arabia

Abstract: The widespread use of computers and smartphones has led to an increase in social media usage, where users can express their opinions freely. However, this freedom of expression can be misused for spreading abusive and bullying content online. To ensure a safe online environment, cybersecurity experts are continuously researching effective and intelligent ways to respond to such activities. In this work, we present ArCyb, a robust machine-learning model for detecting cyberbullying in social media using a manually labeled Arabic dataset. The model achieved 89% prediction accuracy, surpassing the state-of-the-art cyberbullying models. The results of this work can be utilized by social media platforms, government agencies, and internet service providers to detect and prevent the spread of bullying posts in social networks.

Author 1: Khalid T. Mursi

Author 2: Abdulrahman Y. Almalki

Author 3: Moayad M. Alshangiti

Author 4: Faisal S. Alsubaei

Author 5: Ahmed A. Alghamdi

Keywords: Natural language processing; machine learning; neural network; bullying; cyberbullying

Paper 111: Development of a Touchless Control System for a Clinical Robot with Multimodal User Interface

Abstract: This article introduces the development of a multi-modal user interface for touchless control of a clinical robot. This system seamlessly integrates distinct control modalities: voice commands, an accelerometer-embedded gauntlet, and a virtual reality (VR) headset to display real-time robot video and system alerts. By synergizing these control approaches, a more versatile and intuitive means of commanding the robot has been established. This assertion finds support through comprehensive assessments conducted with both seasoned professionals and novices in the domain of clinical robotics, all within a controlled experimental setting. The diverse array of test results unequivocally demonstrate the system’s efficacy. They substantiate the system’s ability to proficiently govern a robotic arm in the clinical environment. The user interface’s usability is measured at an impressive 90.2 on the system usability scale, affirming its suitability for robotic control. Notably, the interface not only offers comfort but also intuitiveness for operators of varying levels of expertise.

Author 1: Julio Alegre Luna

Author 2: Anthony Vasquez Rivera

Author 3: Alejandra Loayza Mendoza

Author 4: Jes´us Talavera S.

Author 5: Andres Montoya A

Keywords: Multimodal user interface; human–robot interaction; clinical robot

Paper 112: Dual-Level Blind Omnidirectional Image Quality Assessment Network Based on Human Visual Perception

Abstract: With the rapid development of virtual reality (VR) technology, a large number of omnidirectional images (OIs) with uncertain quality are flooding into the internet. As a result, Blind Omnidirectional Image Quality Assessment (BOIQA) has become increasingly urgent. The existing solutions mainly focus on manually or automatically extracting high-level features from OIs, which overlook the important guiding role of human visual perception in this immersive experience. To address this issue, a dual-level network based on human visual perception is developed in this paper for BOIQA. Firstly, a human attention branch is proposed, in which the transformer-based model can efficiently represent attentional features of the human eye within a multi-distance perception image pyramid of viewport. Then, inspired by the hierarchical perception of human visual system, a multi-scale perception branch is designed, in which hierarchical features of six orientational viewports are considered and obtained by a residual network in parallel. Additionally, the correlation features among viewports are investigated to assist the multi-viewport feature fusion, in which the feature maps extracted from different viewports are further measured for their similarity and correlation by the attention-based module. Finally, the output values from both branches are regressed by fully connected layer to derive the final predicted quality score. Comprehensive experiments on two public datasets demonstrate the significant superiority of the proposed method.

Author 1: Deyang Liu

Author 2: Lu Zhang

Author 3: Lifei Wan

Author 4: Wei Yao

Author 5: Jian Ma

Author 6: Youzhi Zhang

Keywords: Omnidirectional image quality assessment; dual-level network; human visual perception; human attention; multi-scale

Paper 113: A Novel Voice Feature AVA and its Application to the Pathological Voice Detection Through Machine Learning

Abstract: Voice pathology is a universal problem which must be addressed. Traditionally, this malady is treated by using the surgical instruments in the varied healthcare settings. In the current era, machine learning experts have paid an increasing attention towards the solution of this problem by exploiting the signal processing of the voice. For this purpose, numerous voice features have been capitalized to classify the healthy and pathological voice signals. In particular, Mel-Frequency Cepstral Coefficients (MFCC) is a widely used feature in speech and audio signal processing. It denotes spectral characteristics of a voice signal, particularly of human speech. The modus operandi of MFCC is too time-consuming, which goes against the hasty and urgent nature of the modern times. This study has developed a yet another voice feature by utilizing the average value of the amplitudes (AVA) of the voice signals. Moreover, Gaussian Naive Bayes classifier has been employed to classify the given voice signals as healthy or pathological. Apart from that, the dataset has been acquired from the SVD (Saarbrucken Voice Database) to demonstrate the workability of the proposed voice feature and its usage in the classifier. The machine experimentation rendered very promising results. Particularly, Recall, F1 and accuracy scores obtained, are 100%, 83% and 80%, respectively. These results vividly imply that the proposed classifier can be installed in various healthcare settings.

Author 1: Abdulrehman Altaf

Author 2: Hairulnizam Mahdin

Author 3: Ruhaila Maskat

Author 4: Shazlyn Milleana Shaharudin

Author 5: Abdullah Altaf

Author 6: Awais Mahmood

Keywords: Pathological voice; healthy voice; voice feature; amplitudes; machine learning

Paper 114: Improved Model for Smoke Detection Based on Concentration Features using YOLOv7tiny

Abstract: Smoke is often present in the early stages of a fire. Detecting low smoke concentration and small targets during these early stages can be challenging. This paper proposes an improved smoke detection algorithm that leverages the characteristics of smoke concentration using YOLOv7tiny. The improved algorithm consists of the following components: 1) utilizing the dark channel prior theory to extract smoke concentration characteristics and using the synthesized αRGB image as an input feature to enhance the features of sparse smoke; 2) designing a light-BiFPN multi-scale feature fusion structure to improve the detection performance of small target smoke; 3) using depth separable convolution to replace the original standard convolution and reduce the model parameter quantity. Experimental results on a self-made dataset show that the improved algorithm performs better in detecting sparse smoke and small target smoke, with mAP@0.5 and Recall reaching 94.03% and 95.62% respectively, and the detection FPS increasing to 118.78 frames/s. Moreover, the model parameter quantity decreases to 4.97M. The improved algorithm demonstrates superior performance in the detection of sparse and small smoke in the early stages of a fire.

Author 1: Yuanpan ZHENG

Author 2: Liwei Niu

Author 3: Xinxin GAN

Author 4: Hui WANG

Author 5: Boyang XU

Author 6: Zhenyu WANG

Keywords: YOLOv7tiny; smoke detection; dark channel; smoke concentration; feature fusion; depthwise separable convolution

Paper 115: Virtual Machine Allocation in Cloud Computing Environments using Giant Trevally Optimizer

Abstract: Cloud computing has gained prominence due to its potential for computational tasks, but the associated energy consumption and carbon emissions remain significant challenges. Allocating Virtual Machines (VMs) to Physical Machines (PMs) in cloud data centers, a known NP-hard problem, offers an avenue for enhancing energy efficiency. This paper presents an energy-conscious optimization approach utilizing the Giant Trevally Optimizer (GTO) which is inspired by the hunting strategies of the giant trevally, a proficient marine predator. Our study mathematically models the trevally's hunting behavior when targeting seabirds. The trevally's approach involves strategic selection of optimal hunting locations based on food availability, including pursuing seabird prey in the air or seizing it from the water's surface. Through extensive simulations, our method demonstrates superior performance in terms of skewness, CPU utilization, memory utilization, and overall resource allocation efficiency. This research offers a promising avenue for addressing the energy consumption challenges in cloud data centers while optimizing resource utilization for sustainable and cost-effective cloud operations.

Author 1: Hai-yu Zhang

Keywords: Cloud computing; resource allocation; virtualization; Giant Trevally Optimizer

Paper 116: Corpus Generation to Develop Amharic Morphological Segmenter

Abstract: Morphological segmenter is an important component in Amharic natural language processing systems. Despite this fact, Amharic lacks large amount of morphologically segmented corpus. Large amount of corpus is often a requirement to develop neural network-based language technologies. This paper presents an alternative method to generate large amount of morph-segmented corpus for Amharic language. First, a relatively small (138,400 words) morphologically annotated Amharic seed-corpus is manually prepared. The annotation enables to identify preﬁxes, stem, and sufﬁxes of a given word. Second, a supervised approach is used to create a conditional random ﬁeld-based seed-model (on the seed-corpus). Applying the seed-model (an unsupervised technique on a large unsegmented raw Amharic words) for prediction, a large corpus size (3,777,283) of segmented words are automatically generated. Third, the newly generated corpus is used to train an Amharic morphological segmenter (based on a supervised neural sequence-to-sequence (seq2seq) approach using character embeddings). Using the seq2seq method, an F-score of 98.65% was measured. Results show an agreement with previous efforts for Arabic language. The work presented here has profound implications for future studies of Ethiopian language technologies and may one day help solve the problem of the digital-divide between resource-rich and under-resourced languages.

Author 1: Terefe Feyisa

Author 2: Seble Hailu

Keywords: Amharic; Amharic morphology; segmentation corpus; seq2seq; under-resourced languages

Paper 117: A Novel Fingerprint Liveness Detection Method using Empirical Mode Decomposition and Neural Network

Abstract: One of the most common biometric systems is fingerprint identification, which has been misused due to issues such as fraud. Hence, intelligent methods should be designed and used to recognize real-live fingerprints. Therefore, in the current work, we proposed a novel liveness fingerprint detection framework with low computational cost and excellent accuracy based on empirical mode decomposition and neural network to distinguish real from fake fingerprints. Our proposed scheme works based on empirical mode decomposition technique. The fingerprint images were cropped into 200 × 200 images and then the two-dimensional (2D) images were converted into one-dimensional (1D) data, greatly reducing the computational process. The empirical mode decomposition (EMD) technique decomposed the data and the first five intrinsic mode functions (IMFs) were targeted for feature extraction through simple statistical features. The findings revealed that our suggested system can yield an average accuracy of 97.72% in distinguishing fake from real fingerprints through multilayer perceptron (MLP) neural network. This framework is very efficient compared to other techniques because only one piece of fingerprint image is enough to defend against spoof attacks. Therefore, such framework can reduce the cost of the fingerprint biometric systems, as no further hardware is needed. In addition, our framework method gives the best classification results in comparison to other previous techniques in real-live fingerprint recognition while being simple with lower computational cost. Therefore, this framework can be practically used in commercial biometric systems.

Author 1: Shekun Tong

Author 2: Chunmeng Lu

Keywords: Fingerprint; liveness; biometric; neural network; empirical mode decomposition

Paper 118: Providing a Hybrid and Symmetric Encryption Solution to Provide Security in Cloud Data Centers

Abstract: One of the most crucial components of information technology infrastructure in the modern world is cloud data centers. Customers have access to these data centers' infrastructure and software, which enable them to store and process massive amounts of data. However, the security and protection of private data in cloud data centers is a serious problem that needs effective and c solutions. Security and privacy issues exist because cloud computing outsources the processing of sensitive data. Consumer worries about cloud infrastructure security remain, particularly those related to data privacy. A thorough analysis of research efforts in the area of cloud security is the main objective of this study. In order to do this, a variety of models were evaluated, their advantages and disadvantages were identified, and a viable security solution based on symmetric algorithms was put forth. The original text in the proposed solution (Hybrid encryption algorithm) is first encrypted using the faster symmetric key method AES, and then its key is encrypted using the faster asymmetric key scheme RSA. This increases efficiency and speed. This method will shorten the time required for data encryption while enhancing its security. The final step was implementing the desired solution in the Eclipse software environment and comparing it against the Blowfish and RSA algorithms. The evaluation's findings indicate that the solution is more advantageous, which has resulted in a nearly two-fold decrease in execution time and a marked increase in throughput when compared to the RSA algorithm. Additionally, the execution time has shrunk, and throughput has been vastly improved compared to the Blowfish method.

Author 1: Desong Shen

Keywords: Hybrid encryption algorithm; security; cloud computing; symmetric algorithms

Paper 119: A Single-Stage Deep Learning-based Approach for Real-Time License Plate Recognition in Smart Parking System

Abstract: License plate recognition in smart parking systems plays a crucial role in enhancing parking management efficiency and security. Traditional methods and deep learning-based approaches have been explored for license plate recognition. Deep learning methods have gained prominence due to their ability to extract meaningful features and achieve high accuracy rates. However, existing deep learning-based fire detection methods face challenges in terms of accuracy, real-time requirement, and computation cost, as evident from previous studies. To address these challenges, we propose a single-stage deep learning approach using YOLO (You Only Look Once) algorithm. Our method involves generating a custom dataset and conducting training, validation, and testing processes to train the YOLO-based model. Experimental results and performance evaluations demonstrate that our proposed method achieves high accuracy rates and satisfies real-time requirements, validating its effectiveness for license plate recognition in smart parking systems.

Author 1: Lina YU

Author 2: Shaokun LIU

Keywords: Smart parking; license plate recognition; deep learning; single-stage detector; Yolo

Paper 120: Enhancing Decision-Making with Data Science in the Internet of Things Environments

Abstract: The Internet of Things (IoT) has emerged as a transformative technology, enabling various devices to interconnect and generate vast amounts of data. The insights contained within this data can revolutionize industries and improve decision-making processes. The heterogeneity, scale, and complexity of IoT data pose challenges for efficient analysis and utilization. In this paper, the field of data science is explored in the IoT context, focusing on critical techniques, applications, and challenges vital to realizing the full potential of IoT data. This paper explores the field of data science in the IoT context, focusing on critical techniques, applications, and challenges vital to realizing the full potential of IoT data. The distinctive qualities of IoT data, including its volume, velocity, variety, and veracity, are examined, and their impact on data science approaches is analyzed. Additionally, cutting-edge data science approaches and methodologies designed for IoT data, such as data preprocessing, data fusion, machine learning, and anomaly detection, are discussed. The importance of scalable and distributed data processing frameworks to handle IoT data's large-scale and real-time nature is highlighted. Furthermore, the application of data science in various IoT fields, such as smart cities, healthcare, agriculture, and industrial IoT, is explored. Finally, areas for future research and development are identified, such as privacy and security issues, understanding machine learning models, and ethical aspects of data science in IoT.

Author 1: Lei Hu

Author 2: Yangxia Shu

Keywords: Internet of Things; IoT data; data science; data preprocessing; machine learning; real-time analytics

Paper 121: A Fruit Ripeness Detection Method using Adapted Deep Learning-based Approach

Abstract: Fruit ripeness detection plays a crucial role in precise agriculture, enabling optimal harvesting and post-harvest handling. Various methods have been investigated in the literature for fruit ripeness detection in vision-based systems, with deep learning approaches demonstrating superior accuracy compared to other approaches. However, the current research challenge lies in achieving high accuracy rates in deep learning-based fruit ripeness detection. In this study proposes a method based on the YOLOv8 algorithm to address this challenge. The proposed method involves generating a model using a custom dataset and conducting training, validation, and testing processes. Experimental results and performance evaluation demonstrate the effectiveness of the proposed method in achieving accurate fruit ripeness detection. The proposed method surpasses existing approaches through extensive experiments and performance analysis, providing a reliable solution for fruit ripeness detection in precise agriculture.

Author 1: Weiwei Zhang

Keywords: Fruit ripeness detection; precise agriculture; deep learning; vision system; YOLOv8

Paper 122: A New Method for Intrusion Detection in Computer Networks using Computational Intelligence Algorithms

Abstract: This paper introduces a novel and integrated approach to intrusion detection in computer networks that makes use of the benefits of both abuse detection and anomaly detection techniques. The proposed method combines anomaly detection and abuse detection technologies to enhance intrusion detection functionality. The intrusion detection system is implemented using a set of algorithms and models in the proposed approach. The frog jump algorithm has been utilized to choose the system's ideal input attributes. The decision tree is utilized in this system's abuse detection portion. Support vector machines or basic-radial neural network models have been utilized to find anomalies in this system. In the process of training neural networks, other techniques like particle swarm or genetic optimization are also utilized. The NSL-KDD dataset was used the experiment, and the findings were published. These findings demonstrate that, in comparison to using only anomaly or abuse detection, the proposed approach can increase the effectiveness of intrusion detection in the network. Additionally, a model that uses the frog leap algorithm for feature selection and classification and combines decision tree and support vector machine techniques with ten chosen input features has a detection rate of 98.2%. This is true despite the fact that the detection rates of the systems trained using comparable data in prior studies with 33 and 14 selected input features to the trainer have been 83.2% and 84.2%, respectively. Additionally, the algorithm execution performance increases up to 29 times faster than the aforementioned approaches when the intrusion detection rate is maintained at the level of other competing methods that were simulated in this work.

Author 1: Yanrong HAO

Author 2: Shaohui YAN

Keywords: Decision tree; network intrusion detection; particle swarm algorithm; basic-radial neural network; frog jump algorithm

Paper 123: Autism Diagnosis using Linear and Nonlinear Analysis of Resting-State EEG and Self-Organizing Map

Abstract: The prevalence of autism has increased dramatically in recent years and many people around the world are facing this difficult condition. There is a need to develop an objective method to diagnose autism. Various analysis methods have been used to classify the EEG signals of people with autism, from linear methods in the time and frequency domain to nonlinear methods based on chaos theory. However, there is still no consensus on which method of EEG signal analysis can provide us with the best diagnostic accuracy and valid biomarkers for autism diagnosis. Therefore, in this study, we evaluate different feature extraction methods from EEG signals to diagnose autism from healthy individuals. For this purpose, EEG analysis was performed in time, time-frequency, frequency and nonlinear domains. Furthermore, the self-organizing map (SOM) method was used to classify features extracted from autistic and normal EEG. The data used in this study were recorded by the research team from 24 children with autism and 24 normal children. The accuracies of 92.31, 93.57, 95.63 and 97.10% were achieved through time and morphological, frequency, time-frequency and nonlinear analyzes, respectively. Indeed, the findings showed that nonlinear analysis could yield the best classification results (accuracy = 97.10%, sensitivity = 98.80% and specificity = 97.02%) in the EEG discrimination of autistic children from typical children through the SOM neural network.

Author 1: Jie Xu

Author 2: Wenxiao Yang

Keywords: Autism; EEG; linear analysis; nonlinear analysis; neural network

Paper 124: A Composite Noise Removal Network Based on Multi-domain Adaptation

Abstract: Addressing the limitation of conventional single-scene image denoising algorithms in filtering mixed environmental disturbances, and recognizing the drawbacks of cascaded image enhancement algorithms, which have poor real-time performance and high computational demands, The composite weather adaptive denoising network (CWADN) is proposed. A Cascade Hourglass Feature Extraction Network is constructed with a visual attention mechanism to extract characteristics of rain, fog, and low-light noise from authentic natural images. These features are then transferred from their original real distribution domain to a synthetic distribution domain using a deep residual convolutional neural network. The generator and style encoder of the adversarial network work together to adaptively remove the transferred noise through a combination of supervised and unsupervised training, this approach achieves adaptive denoising capabilities tailored to complex natural environmental noise. Experimental results demonstrate that the proposed denoising network yields a high signal-to-noise ratio while maintaining excellent image fidelity. It effectively prevents image distortion, particularly in critical target areas. Additionally, it adapts to various types of mixed noise, making it a valuable tool for preprocessing images in advanced machine vision algorithms such as target recognition and tracking.

Author 1: Fan Bai

Author 2: Pengfei Li

Author 3: Haoyang Sun

Author 4: Hui Zhang

Keywords: Image denoising; domain adaptation; generative adversarial network; autoencoder