The high rate of increasing latency-sensitive Internet of Things (IoT) application and smart city systems has revealed the constraints of the traditional task scheduling algorithms in the fog-cloud systems, particularly in the situations were dynamically varying workloads and the type of resources may be changed. This paper presents a better Adaptive Task Scheduling system by extending Deep Reinforcement Learning (DRL) based RTAMOTS model and adding deadline feasibility, adaptive multi-objective reward shaping, action masking, and fairness-based scheduling to the model. The suggested structure is superior to the existing DRL schedulers since it takes task viability into account prior to scheduling to optimize the Service Level Agreement (SLA) compliance, whereas a dynamically weighted reward functionalities exchange deadline satisfaction, energy usage, and queue stability in training. Action masking eliminates infeasible scheduling decisions, and priority aging algorithm prevents task starvation, such that resources are fairly exploited on the fog and cloud machine. The large-scale simulations experiments like the comparative analysis with several state-of-the-art scheduling models suggest that the improved RTAMOTS is far more makespan- and energy-efficient and more equitable and compliant with SLA. These results indicate that the proposed solution possesses a scalable and effective solution to real-time tasks scheduling in fog-cloud computing systems.

This review investigates the transformation of deep learning in a fog computing environment, strongly emphasizing synergy between these enabled technologies and their real-world consequences across various domains. Fog computing is the decentralized approach to data processing, overcoming certain limitations in traditional cloud systems: it reduces latency up to 50%, minimizes bandwidth usage, and alleviates network congestion. Deep learning, known for pattern extraction from complex datasets, enhances real-time analytics and intelligent decision-making in resource-constrained environments. Together, they enable effective processing and prompt decision-making in applications such as anomaly detection in healthcare-for example, arrhythmias with 50% faster response, traffic flow optimization in smart cities, and predictive maintenance in industrial automation, reducing downtime by 60%. Integrating deep learning with fog computing has numerous advantages, such as reducing dependencies on cloud infrastructure, enhancing data privacy, and increasing real-time processing. Yet, several challenges remain, like the resource-limited computational capacity of fog nodes, security vulnerabilities, and the need for scalable and efficient architecture. Recent lightweight model design, federated learning techniques, and hierarchical frameworks are some promising solutions to such challenges. This review synthesizes the current research findings, identifies sector-specific applications, and addresses critical challenges. It also outlines future directions comprising the development of adaptive architectures, privacy-preserving methodologies, and hybrid approaches in artificial intelligence. Meeting these challenges will unlock the full potential of deep learning and fog computing-driving innovation and efficiency across industries.

This study combines machine learning (ML) and X-ray imaging to evaluate the health of solder joints in printed circuit boards (PCBs). A convolutional neural network (CNN) served as the base framework, with CNN-LSTM and CNN-CapsNet models added to enhance performance. Pre-training with the CNN facilitated feature extraction, boosting the subsequent performance of LSTM and CapsNet models. The research focused on three objectives: identifying the best ML model for limited datasets, addressing class imbalance in defective solder samples with data augmentation, and using image manipulation to assess model strengths and limitations. Data augmentation significantly improved model accuracies, with CNN, LSTM, and CapsNet achieving 87.05%, 91.29%, and 94.65%, respectively, compared to 76.23%, 83.32%, and 88.05% without augmentation. CapsNet outperformed other models, leveraging its dynamic routing mechanism to preserve feature hierarchies and maintain stable performance. LSTM demonstrated rapid learning through memory cells, while CNNs were prone to overfitting. CapsNet also excelled in balancing classification across solder types, highlighting its ability to handle complex feature relationships. Robustness tests showed CapsNet's resilience to image transformations like rotation, scaling, and flipping, though extreme deformations remained challenging. These results underscore CapsNet's potential for accurate and reliable solder joint classification in diverse scenarios.

This article explores the seven outstanding deep-learning techniques used to enhance network security. It provides a comprehensive analysis of how these techniques address various cybersecurity challenges, including intrusion detection, malware classification, and anomaly detection. This review highlights the effectiveness of deep learning models such as Convolutional Neural Networks (Recurrent neural networks (RNNs) and automatic encoders used in processing large datasets and identifying complex patterns representing security threats. The article also discusses the advantages and limitations of each technique, emphasizing the importance of feature extraction, model training, and real-time processing capabilities. By combining the findings of the current research, this review aims to guide future research and practical implementation of deep learning in securing network infrastructure against evolving cyber threats. The review provided a comprehensive summary of the deep learning techniques used in network security, highlighting their strengths and limitations. The findings showed that deep learning has significant potential to improve detection and response to network threats, although challenges related to model interpretability, data quality, and computational efficiency should be addressed.

Resource allocation has been a very significant topic for both research and development over the last two decades. Given the increasing volume of data, the proliferation of connected devices, and the demand for seamless service delivery, optimal resource allocation has become a vital factor that influences cloud performance. Recently, deep learning-a subcategory of machine learning-seems to possess a great potential to answer this challenge by enabling predictive, adaptive, and self-organized resource allocation. For the first time, this review embraces all the major milestones achieved in dynamic resource allocation with a discussion on over 25+ peer-reviewed articles published from the year 2000 to 2024. This review has emphasized the use of CNNs, RNNs, and other variants of deep learning approaches. Such a review provides a better view of the potential benefits of the different methodologies by highlighting the pros and cons of each. It also covers the use cases, computational methodologies that discuss algorithmic novelty and challenges in scalability, latency, and energy efficiency. A summary of the development in tech was made by comparison in a table to give a meta-view for the top-ten studies. These findings have important implications for cloud service delivery in applications ranging from industrial automation to consumer-oriented applications. They showcase the vast possibilities of deep learning for changing cloud network operations through advanced optimization and point out several open issues, including the integration of federated and edge learning models that will be necessary to achieve improved decentralization and preservation of network information privacy.

Graph neural networks (GNNs) have become a powerful framework for analyzing structured data in the form of graphs, with applications spanning diverse fields such as social networks, biology, and recommender systems. This survey explores methodology, development, and advances in GNN architecture. We methodically survey the major classes of GNNs, including convolutional GNNs (ConvGNNs), spatial–temporal graph neural networks (STGNNs), recurrent-based GNNs (RecGNNs), and graph autoencoders (GAEs). Every model is discussed in terms of underlying mathematical formulations, design principles, and practical applications. This survey aims to provide a comprehensive understanding of GNNs for practitioners, students, and researchers alike, highlighting their versatility and potential for future innovations in graph neural networks. This review is broad, addressing the basic ideas behind GNNs, different architectural designs, training and inference methods, common issues and constraints, the variety of datasets used, and real-world applications across numerous fields. We will furthermore discuss applications of graph neural networks across different fields and exemplify open-source codes, benchmark datasets, and model valuation for graph neural networks. In the end, this survey specifies existing challenges in interpretability, generalization, and scalability and proposes possible future research trends to further promote the performance of GNNs across various graph-based learning missions.

The emergence of 5G networks has revolutionized communication systems by providing unprecedented speed, connectivity, and reliability. This breakthrough technology enables diverse applications such as autonomous vehicles, smart cities, and industrial automation through higher bandwidth and ultra-low latency. However, maintaining consistent Quality of Service (QoS) across these varied applications presents significant challenges due to their conflicting demands. Traditional QoS management methods struggle to address the dynamic and complex requirements of 5G, prompting the adoption of Machine Learning (ML) techniques. ML offers intelligent, adaptive solutions for traffic prediction, network slicing, and real-time decision-making, ensuring improved resource allocation and seamless service delivery.

Dynamic resource management is important for 5G wireless networks to ensure they are efficient, scalable, and can handle growing connectivity demands while maintaining quality service. The aim of this review is to discuss how deep learning has changed the way complex challenges are being addressed in resource allocation, frequency spectrum management, energy efficiency, and runtime decision-making over 5G wireless networks. It combines the very best of leading-edge research insights into showing, through advanced deep learning techniques like supervised learning, and federated learning, how to allow for intelligent, adaptive solutions that go beyond conventional approaches. The manuscript describes this through a review that compares the strengths of these methodologies in network performance optimization while pointing out some limitations related to computational complexity or lack of extensive real-world testing. It further elaborates on promising future directions, ranging from federated learning for decentralized resource management to enhancing the interpretability of deep learning models and leveraging diverse datasets for improving robustness. The discussion also covers the arrival of 6G networks, which will introduce refined and AI-driven approaches for resource optimization. By establishing the logical links between theoretical developments and practical uses, the presented review will pinpoint the transforming potential of deep learning in re-shaping both the wireless communication networks of the future, but also opening new frontiers well beyond 5G.

This paper targets the development of advanced machine learning strategies for fog computing systems and is designed to further enhance current mechanisms related to resource allocation. Fog computing represents the extension of cloud facilities to network edges with increased data processing, allowing minimal latency for applications that need real-time processing. This is a review underlining deep learning as one of the basic tools through which neural networks predict the resource usage and optimization of resource allocation with its dynamic adaptation to modifications within the network conditions. The paper reviews techniques such as Convolutional Neural Networks, Recurrent Neural Networks, and Generative Adversarial Networks that are explored for their roles in enhancing efficiency, privacy, and responsiveness within the realm of distributed environments. These findings reveal that deep learning significantly enhances operational performance, reduces latency, and strengthens security in fog networks. By processing data locally and autonomously managing resources, these strategies ensure efficient handling of diverse and dynamic demands. It concludes that the integration of machine learning into fog computing forms a scalable and robust framework toward meeting modern challenges imposed by digital ecosystems, enabling smarter real-time decision-making systems at the edge.

This review explores the crucial role of fog computing in addressing the increasing demands of the Internet of Things (IoT) and cloud environments, focusing on its ability to reduce latency, manage large data volumes, and optimize bandwidth by bringing computational resources closer to data sources. The paper highlights three key contributions: first, it discusses significant advancements in fog computing infrastructure, such as the development of scalable fog nodes and improved software solutions, which enhance deployment and management capabilities; second, it examines how fog computing is being integrated with related technologies like IoT, 5G, and blockchain to create efficient, secure, and decentralized systems; and third, it addresses the technical challenges associated with scalability, interoperability, and privacy, emphasizing the need for stronger security mechanisms, resource management algorithms, and standardized protocols. The review also looks ahead to future research directions, particularly the potential of artificial intelligence (AI) to optimize fog computing systems and the role of 5G networks in expanding their capabilities. Overall, the article provides a comprehensive overview of the state of fog computing, detailing its current advancements, ongoing challenges, and the opportunities it holds for transforming connected and data-driven systems in the future.

The integration of deep learning (DL) applications with the Internet of Things (IoT) has emerged as a transformative approach for advancing smart healthcare systems. This review synthesizes findings from seven research studies, each exploring the intersection of these technologies in improving healthcare delivery, patient monitoring, and medical decision-making. The paper highlights how IoT devices, including sensors and wearables, generate vast amounts of real-time health data, which DL models leverage for predictive analytics, diagnosis, and personalized treatment recommendations. Key areas explored include: Data Acquisition and Processing: IoT-enabled sensors play a critical role in collecting physiological data, such as heart rate, blood pressure, and glucose levels, which are then processed by DL algorithms to identify patterns and anomalies, Remote Patient Monitoring: The combination of IoT and DL facilitates continuous monitoring of chronic conditions and allows for real-time intervention, reducing hospital readmissions and enhancing patient independence.

This research aims to investigate the application of machine learning (ML) techniques in network anomaly detection to enhance security in the face of evolving cyber threats. Employing a systematic review of existing literature and experimental evaluation, the study explores the effectiveness of various ML algorithms and their capacity to detect anomalies in network traffic. Unlike traditional rule-based methods, ML algorithms analyze extensive traffic data to distinguish normal from abnormal behavior, adapting dynamically to new threats in real-time. Key methodologies include feature engineering to optimize model performance, focusing on attributes like packet size and flow duration. The research evaluates detection accuracy, reduction of false positives, and the adaptability of ML-based systems to changing conditions. Main outcomes demonstrate that ML offers significant advantages over heuristic approaches, with improved detection rates, minimized human intervention, and enhanced responsiveness to emerging threats. The findings underscore the importance of real-time detection capabilities and highlight challenges such as computational complexity and dataset quality. By addressing these challenges, the study contributes valuable insights into strengthening network defense mechanisms through advanced ML applications.

The smart electrical grid represents a significant advancement in generating, distributing, and consuming electricity. This sophisticated system integrates modern technology and communication tools to enhance energy management efficiency and improve demand costuming within the power network. In this paper, optimal operation of the electrical network with energy management and Demand Response Program (DRP) is implemented. The implementation of the optimal operation is done via multi-stage and multi-objective functions modeling. The DRP modeling is done in first stage to optimal management of consumption in demand side. In second stage, operating cost, emission, power losses and voltage profile are optimized as multi-objective functions modeling with attention to optimal management of consumption in demand side. The solving optimal operation of the electrical network is carried out by using Elephant Herding Optimization (EHO). This problem is implemented on 33-bus test system with hybrid energy resources. Finally, DRP leads to reducing costs, emissions and losses and improving voltage profile in proposed electrical network. Hence, operation costs, emission, power losses, and voltage deviation with the participation of DRP are minimized by 39.15%, 9.94%, 33.35%, and 30.73%, respectively. On the other side, voltage stability is enhanced by 3.66% without considering DRP.

This paper introduces a prediction-based interference algorithm in a multi-UAV scenario to calculate and apply the minimum interference-free flight time. The algorithm summarizes the UAV dynamic equation into kinematic equations using a low-level controller. Neighboring UAVs exchange their predicted trajectories at each sampling time to predict interferences. Then, the distributed nonlinear predictor under the interference resolution law and strategy predicts the possible control variables for each UAV and calculates the minimum required travel time. Subsequently, a collision avoidance system is designed based on a predictive distributed controller for tracking, in which anti-collision constraints are defined according to the International Civil Aviation Organization (ICAO) priority rights. To reduce the computational burden, the predictive distributed controller is formulated as a quadratic integer programming optimization problem. The results show that the proposal can resolve conflicts in real time and in the presence of a crowded airspace, while there is no interference and secondary interference. The proposed algorithms were simulated using MATLAB software and the estimated time was compared with the flight time, which showed the favorable performance of the proposed algorithm. Also, to verify the effectiveness of the interference solution, the deviation of the aircraft flight path due to maneuvers and the increase in the path length were shown, and the value of the collision avoidance system was more accurate compared to other aircraft with less efficiency.

BACKGROUND:Machine learning could be used for prognosis/diagnosis of maternal and neonates' diseases by analyzing the data sets and profiles obtained from a pregnant mother. PURPOSE:We aimed to develop a prediction model based on machine learning algorithms to determine important maternal characteristics and neonates' anthropometric profiles as the predictors of neonates' health status. METHODS:This study was conducted among 1280 pregnant women referred to healthcare centers to receive antenatal care. We evaluated several machine learning methods, including support vector machine (SVM), Ensemble, K-Nearest Neighbor (KNN), Naïve Bayes (NB), and Decision tree classifiers, to predict newborn health state. RESULTS:The minimum redundancy-maximum relevance (MRMR) algorithm revealed that variables, including head circumference of neonates, pregnancy intention, and drug consumption history during pregnancy, were top-scored features for classifying normal and unhealthy infants. Among the different classification methods, the SVM classifier had the best performance. The average values of accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC) in the test group were 75%, 75%, 76%, 76%, and 65%, respectively, for SVM model. CONCLUSION:Machine learning methods can efficiently forecast the neonate's health status among pregnant women. This study proposed a new approach toward the integration of maternal data and neonate profiles to facilitate the prediction of neonates' health status.

Building upon the sociotechnical system theory, the present study contributes by examining the relationship between artificial intelligence (AI) adoption, employees' innovative behaviour on employee performance and job security (JS). The primary data is collected from 340 employees from firms located in the industrial hub of a developing economy using a simple random technique, and data is analysed using Smart-PLS 3 from the manufacturing sector. The study evidences that employees' adoption and utilization of AI technologies positively influence their innovative behaviour, job performance (JP), and security. Moreover, the study finds a mediating role of innovative behaviour to connect the dots. Organizations can prioritize using AI-driven training programmes so employees can use AI tools efficiently. Study findings also encourage employees to engage in innovative work behaviours like investigating novel concepts and experimenting with AI technologies to improve JP. This study invalidates that AI will replace employees at the workplace, as we can safely conclude that AI adoption enhances JP and JS.

This study aims to investigate and optimize the energy usage of multiple electrical systems operating off-grid. The focus is on maximizing the utilization of storage systems to ensure a reliable power supply. This is achieved by harnessing renewable energy sources such as backup storage systems, wind turbines, solar cells, and diesel generators. To meet the energy demand, both DC and AC loads on the consumer side are considered. The research employs a mathematical approach known as nonlinear quadratic programming to ascertain the most efficient energy generation for both producers and consumers. The primary aim is to minimize expenses the system. The study also evaluates the impact of battery storage systems on achieving the operational level. By analyzing the performance of these storage systems, the researchers can assess their contribution to the overall energy optimization. To validate the proposed optimization algorithm, the study implements it in Matlab software. This enables numerical simulations in various case studies of energy systems demonstrates the optimal energy generation and offers valuable insights into the performance of the proposed algorithm. Overall, this study contributes to the field of energy optimization in off-grid systems by considering multiple electrical systems, energy sources, and storage systems. The findings can help in designing and operating off-grid systems more efficiently, ensuring a reliable and cost-effective power supply.

This study proposes day-ahead power scheduling for electrical systems in off-grid mode, emphasizing consumer involvement. Bi-Demand Side Management (DSM) approaches like strategic conversion and demand shifting are proposed for consumer involvement. Multiple objectives are modelled to voltage profile improvement and reduce the operation energy cost. The non-dominated solutions of the voltage of buses and operation energy cost are generated by enhanced epsilon-constraint technique, simultaneously. The General Algebraic Modeling System (GAMS) software is proposed for solving optimization problems. A combination of decision-making methods like weight sum and fuzzy procedures are implemented for finding optimal solution non-dominated solutions. The proposed method’s effectiveness is confirmed through numerical simulations carried out on several case studies that utilize the 33-bus electrical system. The findings illustrate the substantial effectiveness of demand-side participation in improving power dispatch and the optimal rate of multiple objectives. By using DSM, operation cost is reduced by 21.58% and the voltage index is improved by 13.36% than the lack of implementing DSM.

This study aims to enhance the detection and characterisation of anomalies in manufactured parts by integrating machine learning (ML) with resonance frequency spectra data. A key contribution of this work is the development of a novel Impulse Excitation Technique (IET)-based method that effectively evaluates material health and identifies subtle defects by leveraging numerous mathematical and physical metrics as input features. Three machine learning models - Random Forest (RF), K-Nearest Neighbor (KNN), and Multi-layer Perceptron (MLP) - were systematically compared to determine the most effective method for classifying defects, specifically focusing on healthy, cracked, and dimensionally deviated samples. Among these, the MLP model demonstrated the highest performance, achieving Receiver Operating Characteristic (ROC) values of 0.963, 0.901, and 0.942 for each class, respectively. Additionally, SHAP (SHapley Additive exPlanations) analysis showed anomalies were sensitive to specific resonance frequency metrics, improving prediction accuracy. Cracked samples exhibited slight peak broadening and negative peak shifts, while dimensionally deviated samples showed positive and negative shifts and missing peaks. Dimensional deviations were more pronounced than cracks, making them easier to identify and enhancing predictive accuracy.

Given the increasing need for interactive human-computer applications, the field of employing machine learning algorithms to discern emotions from speech has seen a substantial surge in interest. While emotion recognition systems have made substantial progress in languages like German, English, Spanish, Dutch, and Danish, the availability of comprehensive datasets for the Kurdish language remains notably limited. This paper addresses this gap by focusing on emotion recognition in Sorani Kurdish dialect speech data, which was carefully gathered from openly available videos from the YouTube platform and categorized into four clear supposed emotions: neutral, sadness, happiness, and anger. The study applied both natural Mel Spectrogram and Mel-Frequency Cepstral Coefficient (MFCC) features for various spectrals, followed by the classification models K-Nearest Neighbor (KNN), Multi-Layer Perceptron (MLP), and Support Vector Machine (SVM) to evaluate the results. By closely examining and contrasting the results of using several methods for feature extraction, it was found that SVM obtained a higher accuracy, reaching as much as 85.57%. This is so much more than the first Kurdish emotion classification technique for the recognition of the emotion of the words.