Security in public networks has always been a challenge. With an ever-expanding landscape of cyberattacks, it is imperative to explore new/alternate mechanisms of implementing intrusion detection. Machine learning and deep learning techniques have emerged as promising for intrusion detection in the recent past, with increased efficiency. This study explores machine learning algorithms, namely Random Forest, Naive Bayes, and Decision Tree, for the detection of web-based attacks in public networks, using a combination of Principal Component Analysis and Haar Wavelet Transform for feature extraction. The results of these models are compared, and related issues and approaches to alleviate them are explored.
The research paper introduces a quantitative comparison of two reactive power compensation methods of improving the power factor (PF) performance in institutional buildings: automated capacitor banks and synchronous condensers at Universiti Teknikal Malaysia Melaka (UTeM). Three academic buildings, namely the Faculty of Electrical Technology and Engineering (FTKE), the Faculty of Electronic and Computer Technology and Engineering (FTKEK), and the Faculty of Information and Communication Technology (FTMK), were measured with a three-phase power quality analyzer that met the IEC 61000-4-30 Class A standards at 400 V with a frequency of 50 Hz. The measurement results of active power, reactive power, and PF profiles were set into a MATLAB/Simulink simulation model to assess reactive power compensation under realistic operating conditions. The power factor improvement indicator (ΔPF) was used to measure technical performance with the help of the reactive power sizing sensitivity analysis. The regulated requirement of a PF correction target of 0.85 was applied to prevent the penalties on utility surcharges according to Tenaga Nasional Berhad (TNB) regulations. The results of the simulations suggest that both compensation methods are effective in enhancing the PF compared to the utility benchmark, with the system with a low baseline PF having a greater correction sensitivity. Automated capacitor banks offered a cost-efficient response to steady-state compensation, whereas synchronous condensers offered better dynamic reactive response power support and better voltage stability under varying load conditions. The results indicate that effective Power Factor Correction (PFC) strategies can significantly reduce reactive current flow, minimize distribution losses, and improve the utilization of real power, thereby providing a technically and economically viable approach for sustainable energy management in institutional electrical distribution networks.
Smart grid technologies have evolved rapidly to enable intelligent monitoring, forecasting, and control of modern energy systems. However, the increasing integration of renewable energy sources, variable demand, and dynamic grid conditions poses significant challenges for effective energy management. Traditional methods are often not well-suited to handling the nonlinear and time-varying nature of smart grid data, which can lead to reduced prediction accuracy and scheduling efficiency. To address these challenges, this paper proposes a hybrid Sine Cosine Optimization (SCO)–Bidirectional Long Short-Term Memory (BiLSTM)–Deep Reinforcement Learning (DRL) framework for smart grid energy management. The BiLSTM model is employed to accurately forecast energy demand, whereas the SCO algorithm optimizes model parameters to enhance prediction performance. Experimental results show that the proposed model achieves lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) values and higher prediction accuracy than conventional models, including Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and BiLSTM. The proposed framework improves energy utilization efficiency, reduces operating costs, and enhances electrical grid stability. Overall, the proposed approach provides a scalable and efficient solution for next-generation smart grid energy management.
Artificial Intelligence (AI) has shown significant potential in healthcare, particularly for the early detection of glaucoma. This study presents a comprehensive evaluation of glaucoma classification using the ACRIMA dataset. Handcrafted descriptors, such as Edge Histogram (EH), Fuzzy Color and Texture Histogram (FCTH), and Pyramid Histogram of Gradients (PHOG), are combined with deep features extracted from ConvNeXt and Vision Transformer (ViT) models, and their classification performance is evaluated using three different classifiers, namely Random Forest (RF), Gradient Boosting (GB), and LightGBM (LGBM). Four ensembling frameworks, namely Feature-level (FL), Hybrid Feature-level (HF), Voting-based (VB), and Stacking ensembles, are investigated to integrate handcrafted and deep features effectively. Among the proposed approaches, the stacking ensemble using Logistic Regression (LR) as a meta-classifier outperformed other approaches by providing an accuracy of 95.44404% with an F1-score of 96.0638% and an AUC of 0.9887.
This paper proposes a current control strategy for a single-phase grid-tied inverter under conditions of parameter uncertainty and disturbance. Inductance and filter resistance deviations degrade the performance of traditional model-based control methods. To overcome this, a Radial Basis Function (RBF) neural network is used as an online observer to approximate and compensate for aggregate errors due to uncertainty and unmodeled dynamics. Simultaneously, Super-Twisting Sliding Mode Control (STSMC) is applied to reduce the chattering phenomenon and ensure finite-time convergence with a continuous control signal. The RBF–STSMC combination reduces conservatism in parameter selection and enhances system robustness. Stability is demonstrated using Lyapunov, and MATLAB/Simulink simulation results show superior performance compared to traditional SMC and PID controllers.
Orthographic projection is one of the most demanding topics in technical drawing because students must mentally connect three-dimensional forms with their two-dimensional representations. Although mobile Augmented Reality (AR) offers promising support for this type of spatial learning, little is known about how students in technical secondary education perceive and accept such tools, especially in non-Western contexts. This study explored the acceptance of a mobile AR application used for orthographic projection learning by 60 first-year baccalaureate students in electrical and mechanical technology tracks at a Moroccan technical high school. Drawing on the Technology Acceptance Model 2 (TAM2), data were collected using a seven-point Likert scale after an eight-week instructional sequence (16 h in total) during which students used the application in class. Responses were analyzed with SmartPLS 4 using Partial Least Squares Structural Equation Modeling and 5,000 bootstrap subsamples. The measurement model showed strong reliability and convergent validity, with Cronbach's alpha values above 0.88, composite reliability above 0.91, and Average Variance Extracted (AVE) above 0.73. Discriminant validity was also confirmed, with the Heterotrait-Monotrait Ratio (HTMT) values below 0.90. The structural model accounted for 65.9% of the variance in perceived usefulness and 68.5% of the variance in behavioral intention. Perceived usefulness had the strongest effect on students' intention to use the application. Perceived ease of use and job relevance contributed significantly to perceived usefulness, whereas subjective norm, output quality, and result demonstrability did not. Overall, the findings suggest that students' acceptance of AR in technical drawing depends less on social influence or technical features alone than on whether the tool is easy to use and clearly supports classroom tasks. The study extends TAM2 to a specific AR learning situation in Moroccan technical secondary education, a context that remains underrepresented in the literature.
This study examines the impact of adding Polypropylene (PP) fibers on the physical and mechanical properties of lightweight bricks made from pumice sand, using Portland cement as a binder, with a ratio of 1 part cement to 5 parts sand, and concentrations of 0.25%, 0.50%, 0.75%, 1.00%, and 1.25% by weight of cement, including a control specimen without fiber. The results of compressive strength, density, and water absorption tests indicate that adding PP fibers at an optimal level of 0.50% increased compressive strength by approximately 2%, reduced water absorption by about 12%, and improved resistance to microcracking compared to control specimens.
Ultrasonic Welding (UW) is used to assemble additively manufactured thermoplastic parts in applications such as robotics, automotive components, and custom mechanical systems. This study investigates permanent joints of thermoplastic components manufactured by Fused Deposition Modeling (FDM) and assembled using UW. The objective of this work is to determine the optimal joint interface geometry with a 1 mm contact zone thickness and to evaluate the influence of the type of thermoplastic on the mechanical strength of ultrasonic welds. Composite specimens were fabricated via FDM using Acrylonitrile Butadiene Styrene (ABS), Polylactic Acid (PLA), and Polyethylene Terephthalate Glycol (PETG), and were designed with five different Energy Director (ED) geometries: circular, curved, rectangular, keyway-type, and regular hexagonal profiles. Mechanical strength tests were performed using a computer-controlled tensile testing machine at a crosshead speed of 5 mm/min, with the ultimate failure load recorded for each configuration. The microstructure of the welded joints was analyzed using an optical microscope (ZEISS AXIO VERT A1) to assess melt uniformity and interfacial bonding quality. The results demonstrate that the hexagonal ED geometry provides the most uniform melt distribution among the tested configurations. Among the materials investigated, PLA specimens welded ultrasonically exhibited the highest tensile strength, reaching failure loads of up to 3.9 kN, which is significantly higher than those of ABS and PETG. At the same time, PETG demonstrates consistently high strength values (up to 2.92 kN), outperforming ABS in most configurations. Specimens were fabricated via FDM with a 0.2 mm layer height and 80% infill using ABS, PLA, and PETG.
Face recognition with occlusion remains a challenging issue for biometric authentication systems in real-world scenarios. Recent Generative Adversarial Network (GAN)- based approaches have improved facial reconstruction under partial occlusion; however, recognition accuracy remains severely limited in regions with extensive facial occlusion. To address this limitation, this study proposes a multimodal biometric framework, Complete Face Recovery (CFR)-GAN++. The framework combines self-supervised 3D face reconstruction with physiological biometrics from Electroencephalography (EEG) and Electrocardiography (ECG). The proposed framework consists of a facial reconstruction generator based on U-Net, a CNN–BiLSTM EEG encoder, and a 1D CNN ECG encoder in an adaptive feature-level fusion framework. The visual stream reconstructs occlusion-corrupted facial regions with a self-supervision strategy of Swap-Rotate-and-Render and 3D Morphable Model (3DMM) regression.
Smart city traffic management is a multi-attribute decision-making problem where city authorities continuously evaluate alternative intersections under dynamic conditions. Uncertain, interdependent criteria, such as traffic density (e.g., 1200–2800 vehicles/hour), accident risk (5–15% variance), pedestrian flow, weather, pollution, and emergency access, render traditional models inadequate. This study presents a dynamic hypersoft set-based framework for selecting the best traffic management zone using two algorithms. A dynamic choice matrix is constructed, and alternatives are ranked using scalar DHSM values and dynamic choice vectors in Algorithm 1, whereas in Algorithm 2, dynamic value, utility, and score matrices are developed to obtain a final ranking of alternatives. The study demonstrates how the Dynamic Hypersoft Set (DHSS) theory can support adaptive and transparent decision-making in smart city traffic systems.
In modern enterprises, software development is an important part, as businesses are increasingly relying on it to support and improve operational performance. Nevertheless, low-quality software decreases the level of satisfaction of clients. Software development is now considered a combined and technology-dependent activity, and the developer's experience can play an essential role in shaping the quality of software they generate. In software development, the allocation of the appropriate software developers (for example, those who have appropriate coding skills) to a project is an important feature. The difficulty lies in the fact that it is very complicated for project managers, clients, and software development companies to find a suitable developer to allocate a specific project. Therefore, there is a need for a scalable mechanism to identify the level of coding expertise of the software developer. Deep Learning (DL) methods have been extensively applied to assess the impact of developers' experience on code quality. This study presents an Empirical Evaluation of Developer Experience-Based Software Quality Estimation Using Spiking Neural Networks and Metaheuristic Optimization (EESQA-DELMOA) model, which aims to assess software quality by analyzing the developers' experience levels on code reliability and maintainability. EESQA-DELMOA employs a Bio-inspired Artificial Hummingbird (BAHB) technique to select the most relevant features for improving model performance. Subsequently, a Simplified Spiking Neural Network (SSNN) is deployed for classification. Finally, parameter tuning is performed using an Adaptive Migration Butterfly Optimization Algorithm (AMBOA) to improve the classification performance of the SSNN classifier. EESQA-DELMOA was experimentally evaluated using a benchmark dataset, and the results demonstrate its enhanced performance compared to recent approaches.
This study presents an integrated approach to recycling Electric Arc Furnace (EAF) waste within a circular economy. EAF waste, including EAF slag, Ladle Furnace (LF) slag, aspiration dust, and Charge Preparation Shop (CPS) waste, was analyzed using the XRF and XRD methods. The predominance of iron oxide and calcium-magnesium-silicate phases was established, confirming the feasibility of secondary iron recovery and the use of slag in construction materials. Reductive smelting of ore-coke briquettes yielded a metallized product with an iron content of 85%–90%, as well as "iron-free" slag, which was used as a filler for structural concrete, along with LF slag. The most effective composition of ore-coke briquettes contained 45% waste from the CPS, 23% aspiration dust, 23% EAF slag, and 9% coke breeze. The compressive strength of the structural concrete ranged from 20.1 MPa to 33.6 MPa. The proposed process flow diagram ensures the comprehensive recycling of solid waste from electric steelmaking, reducing waste disposal volumes and producing in-demand metallurgical and construction materials within a unified circular economy process chain.
Diesel-powered irrigation pumps are widely used in dryland agriculture but impose high fuel costs and carbon emissions. Solar Photovoltaic (PV) systems offer a cleaner alternative, but their performance is limited under fluctuating irradiance. Hybrid solar–diesel systems can improve stability, yet most existing studies focus on simulations or economic analyses rather than integrated adaptive control with physical implementation. This study develops and evaluates an adaptive solar–diesel hybrid irrigation system integrating PV generation, environmental sensing, an automatic transfer switch, and a predictive control algorithm based on an Artificial Neural Network optimized using a Genetic Algorithm (ANN–GA). The ANN predicts short-term solar irradiance using temperature and humidity inputs, while the GA optimizes model parameters to support real-time energy-source switching. A laboratory-scale prototype (100 Wp PV, 100 Ah battery, 0.5 HP pump) was tested under naturally fluctuating weather conditions over four days. The ANN–GA model achieved high predictive accuracy (coefficient of determination (R²) = 0.9481 for training; R² = 0.9042 for validation), with low prediction errors (Root Mean Square Error (RMSE) = 11.58 W m-2 and Mean Absolute Error (MAE) = 4.70 W m-2), thereby enabling stable, oscillation-free switching decisions. The hybrid system supplied 52% of total energy from PV, reducing diesel runtime by 52%, fuel consumption by 53%, and CO₂ emissions by 53% compared to a diesel-only system. Sensitivity analysis under low irradiance (<200 W m-2) revealed reduced efficiency, emphasizing the need for enhanced storage or complementary renewable sources. These results demonstrate that predictive control integrated with hybrid energy systems can significantly improve irrigation reliability, energy efficiency, and sustainability in tropical dryland environments. This study presents a fully implemented and experimentally validated framework for intelligent, energy-efficient irrigation systems that goes beyond simulation-based approaches.
Oral squamous cell carcinoma (OSCC) is a major global health concern, particularly in low- and middle-income countries such as India, where tobacco and areca-nut use are prevalent. Early detection of oral potentially malignant disorders (OPMDs) greatly improves survival, yet conventional diagnostic methods, such as visual inspection and biopsy, are limited by subjectivity, invasiveness, and resource demands. The integration of artificial intelligence (AI) and digital image processing (DIP) has emerged as a promising solution for non-invasive, scalable screening. This narrative review synthesizes recent literature to provide a thematic overview of digital image processing and artificial intelligence approaches for oral cancer detection, rather than conducting a systematic or meta-analytic evaluation. This review analyzes recent research published on DIP techniques such as noise reduction, contrast enhancement, normalization, and segmentation, and their influence on AI-based classification across photographic, histopathological, and optical coherence tomography (OCT) images. Recent studies demonstrate that preprocessing consistency, hybrid feature extraction, and explainable AI improve diagnostic reliability and interpretability. However, most models remain constrained by small datasets, a lack of external validation, and limited feasibility for deployment. This paper identifies the need for standardized, end-to-end frameworks that integrate preprocessing, feature extraction, and classification to advance AI-based oral cancer detection from experimental feasibility to clinical reality.
This work introduces ENCSNet, an Explainable NASNet-enhanced model based on a multiclass skin image, to improve skin cancer classification. ENCSNet starts feature extraction with NASNet and augments it with a skin cancer-specific CNN. Error Level Analysis (ELA) was utilized as a preprocessing step to detect picture compression differences and improve image input quality. Sequential image augmentation was also employed to escalate feature extraction, so the model is resilient to skin lesions and illumination variations. A focal loss function was used during model training to address class imbalance. Grad-CAM visualization explained how the model makes a judgment by highlighting significant areas for skin cancer diagnosis. ENCSNet performed well on ISIC 2019 with an overall accuracy of 94.65%, a precision of 94.5%, a recall of 94.5%, and an F1-score of 94.5%. The Receiver Operating Characteristic (ROC) analysis of each class also differentiated skin cancer widely. Overall, ENCSNet improves classification accuracy and model transparency compared to other methods, which helps clinical practice diagnose skin cancer using skin scans.
In practice, benzene production is carried out in complex technological facilities such as Benzene Production Units (BPUs). These units include several interconnected subsystems of oil refineries, characterized by several interacting operational parameters that affect both the production volume and the quality indicators of the final product. As a result, the control problem is inherently multi-criteria and is further complicated by operational constraints and uncertainty in part of the input information. In this study, the problem of multi-criteria optimization of the benzene production process control under uncertainty is considered. An efficient heuristic optimization method based on fuzzy modeling is proposed for decision-making in a fuzzy environment, which aims at improving production efficiency while taking into account several conflicting criteria. The validation results obtained through simulation using real operational data from the Atyrau Oil Refinery show that, compared with a deterministic baseline method, the proposed approach increases benzene production by 2.1 t/day (approximately 1.6%), while also outperforming the fuzzy Analytic Hierarchy Process (AHP) method. The obtained results confirm the effectiveness of the proposed approach for supporting decision-making in the management of technological processes under uncertainty.
This study examines how Green Innovation (GI) affects the decoupling of economic growth from CO2 emissions using a multi-country panel dataset for 2012-2023. The analysis extends the Environmental Kuznets Curve (EKC) framework by including GI as both a direct explanatory variable and a moderator in the growth-emissions nexus. Fixed-effects estimation with Driscoll-Kraay standard errors was used to account for heteroskedasticity, serial correlation, and cross-sectional dependence. The results indicate an inverted-U-shaped relationship between income and emissions, but the estimated turning points are not uniformly distributed across the observed income range. GI significantly reduces emissions in high-income countries. In contrast, its effect remains weak in middle- and low-income groups, where absorptive capacity, institutional quality, and energy structure constrain its environmental effectiveness. The interaction term shows that GI attenuates the emissions-intensive effect of growth. The findings imply that innovation policy must be differentiated by development stage rather than treated as a universal decoupling instrument.
Rail surface defects such as flaking, spalling, and squat are critical indicators of railway track degradation and require reliable and efficient automated inspection systems to ensure operational safety. This study presents a comparative evaluation of seven state-of-the-art pre-trained Convolutional Neural Network (CNN) architectures for rail surface defect classification, namely Inception-V3, MobileNet-V1, MobileNet-V2, MobileNet-V3, NasNetMobile, ResNet50, and EfficientNet-B0. Transfer learning was employed by freezing the convolutional backbones and optimizing the classifier head using Optuna-based hyperparameter search. The models were trained and evaluated on a benchmark rail surface defect dataset containing balanced samples of the three defect classes after redundancy reduction and class equalization. Performance was assessed using quality metrics, whereas computational efficiency was analyzed using Floating-Point Operations (FLOPs) and parameter complexity. Experimental results show that MobileNet-V2 achieves the highest classification accuracy of 81.6% with 599M FLOPs, whereas MobileNet-V3 achieves competitive accuracy of 78.5% with the lowest computational cost of only 132M FLOPs. Inception-V3 also demonstrates strong performance with 80.8% accuracy but requires substantially higher computational complexity (11.4B FLOPs). Pareto analysis confirmed that both MobileNet variants provide the best efficiency–accuracy trade-off among all evaluated models. To improve interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to the Pareto-optimal models. The visualizations revealed that MobileNet-V2 generally focuses more consistently on defect-relevant regions, particularly for flaking defects. The findings highlight the importance of combining performance benchmarking with Explainable Artificial Intelligence (XAI) analysis for safety-critical railway monitoring applications.
Accurately forecasting student academic performance enables educational institutions to identify students eligible for scholarships, allocate resources effectively, and design targeted intervention strategies. This study leverages a large-scale longitudinal dataset from King Abdulaziz University to predict students' Grade Point Average (GPA) using machine learning techniques. Prior to modeling, GPA values were normalized to the range of (0–1) to ensure consistency and stability in regression analysis. Two supervised learning models—Support Vector Regression (SVR) and Linear Regression (LR)—were evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results show that both models achieved comparable predictive performance, with SVR slightly outperforming LR. Specifically, SVR achieved an R² of approximately 0.6054, RMSE of 0.5819, and MAE of 0.4476, whereas LR achieved an R² of approximately 0.6046, RMSE of 0.5825, and MAE of 0.4444 when using prior GPA as the primary predictor. The inclusion of demographic variables (gender, marital status, and diploma type) resulted in only marginal improvements, with no statistically significant enhancement in predictive accuracy. These findings confirm that prior academic performance remains the most reliable predictor of future GPA, whereas demographic factors contribute minimally in this context. The proposed models provide an interpretable and reproducible baseline framework that can support scholarship selection processes and the early identification of at-risk students in higher education institutions.
Fire and smoke constitute critical hazards in industrial and urban environments, necessitating reliable early detection systems capable of real-time operation. Although deep learning-based object detection methods have demonstrated considerable potential, existing fire and smoke detection studies predominantly apply default training configurations or modify network architectures, while the systematic investigation of fine-tuning optimization, specifically the interaction effects among optimizer selection, layer freezing depth, learning rate policy, and domain-targeted augmentation, remains insufficiently explored for dual-class fire and smoke detection. This paper proposes a structured four-component fine-tuning framework for the YOLOv9c architecture that jointly optimizes: Stochastic Gradient Descent (SGD) with momentum as a generalization-favoring optimizer, selective freezing of the first 10 backbone layers to preserve transferable features, cosine annealing with linear warmup for controlled domain adaptation, and a fire- and smoke-targeted augmentation pipeline calibrated to the visual variability of fire and smoke phenomena. The key contribution lies not in the individual techniques, which are established in the literature, but in the systematic ablation-based quantification of their individual and interactive contributions, an analysis absent from prior work in this domain. Experiments are conducted on 35,000 manually annotated images sourced from public repositories and Kazakhstan industrial surveillance facilities. The fine-tuned model achieves 78.0% precision, 76.0% recall, and 77.0% F1-score at 30.2 frames per second (FPS), outperforming Faster R-CNN, ResNet-based, YOLOv8m, and YOLOv10m baselines. Ablation analysis reveals that SGD with momentum yields the largest individual improvement (+4.2% F1-score), with the cumulative improvement (+6.3%) exceeding the sum of individual contributions, indicating positive interaction effects among the optimization components.
