In this paper, we introduce a comprehensive Arabic Sign Language (ArSL) recognition system targeting people with hearing disability to bridge the communication gap. Through semantic analysis and AI-driven optimization, we mitigate the challenge of not correctly recognizing sentences and optimizing computational efficiency. Gradient-based adaptive learning (GBL), hyperparameter tuning, and metaheuristic algorithms are integrated to optimize convergence of training and feature extraction to enhance computational efficiency. Automatic hyperparameter selection methods help for adaptive learning rates, leading to better performance of the model without excessive manual involvement. Such AI-driven optimizations result in lower processing overhead while achieving high accuracy while recognizing content. The methodology employs pre-trained transformer models (best practices for BERT and GPT), leading to strong contextual understanding and accurate recognition of full sentences in ArSL. By employing quantization-aware training and optimizing the pruning of models, computational improvements lead to significant memory consumption reductions of 40% and training time reduction of 29%, confirming the compatibility for resource-constrained environments. Comparison of different optimization methods shows that metaheuristics model configurations, e.g., Bayesian Optimization and Genetic Algorithms, present computational trade-offs, validating the choice of the current model configuration. The hardware adaptability is supported by implementation of low-power processing methods that make the system deployable on embedded edge devices. Performance of the system in various datasets on 100,000+ samples is reported to be state-of-the-art with 91% accuracy and 94% F1-score. By loading data incrementally, the model optimizes real-time execution as it is trained on the same dataset, resulting in faster inference times and less latency. At this stage, batch normalization and early stopping are also key to improving computational efficiency, leading to a decrease in training time of 19%. From performance points of view, the system can maintain a high frame processing rate and it has low latency in applications outside the simulation environment. The system also accommodates regional accents and heterogeneous data conditions-signifying scalability and applicability into education, public services and the industry. By filling practical gaps in ArSL recognition, this work demonstrates a robust, effective and inclusive methodology allowing easy transfer with a more reliable system in a realistic scenario and scalable future progress.

Background/Objectives: Obstructivesleep apnea (OSA) is one of the most widespread forms of sleep disease, affecting over 936 million adults globally. The health consequences of obstructive sleep apnea (OSA) are well documented; however, it remains largely underdiagnosed because the current gold-standard diagnostic method, polysomnography (PSG), is often costly, time-consuming, and unavailable in many healthcare settings. To address these challenges, this study presents a novel explainable deep learning (DL) framework for automated multi-class OSA severity classification using single-lead electrocardiogram (ECG) signals. Methods: The proposed framework integrates a hybrid CNN–BiLSTM architecture with explainable artificial intelligence (XAI) techniques to generate clinically meaningful predictions and explanations across four OSA severity classes: Normal, Mild, Moderate, and Severe. The framework was evaluated using the publicly available PhysioNet Apnea-ECG dataset (70 recordings) together with an institutional ECG dataset (150 recordings), resulting in a combined cohort of 220 recordings. Results: The proposed framework achieved an overall classification accuracy of 94.7%, with sensitivity and specificity values of 92.3% and 96.1%, respectively. Furthermore, the proposed model consistently outperformed conventional machine learning algorithms, including Support Vector Machine (SVM), Random Forest, and XGBoost, by 5.5%, 4.2%, and 2.9%, respectively. To enhance transparency and clinical trust, SHAP (SHapley Additive exPlanations) was employed to identify the most influential physiological predictors driving model decisions. Heart rate variability features, particularly RMSSD and pNN50, emerged as the strongest indicators of OSA severity. Moreover, computational efficiency analysis revealed that the model required only 0.23 s to process a 60 s ECG epoch on a standard computing platform, supporting its suitability for real-time deployment. Conclusions: The findings demonstrate that explainable deep learning applied to ECG signals can provide accurate, interpretable, and computationally efficient assessment of OSA severity. The proposed framework may support OSA screening, clinical triage, and early intervention, particularly in resource-constrained healthcare environments.

Mathematical modeling of epidemics is a cornerstone in the study and response to the spread of diseases and related processes across various domains. However, classical models generally do not describe such memory effects properly and are computationally inefficient, which restricts their applicability or predictive accuracy. To address these issues, we introduce a new approach to epidemic modeling using our newly proposed fractional-order differential equations, which are endowed with the Atangana–Baleanu system to describe long-range dependencies and nonlinear characteristics more accurately than the traditional Caputo system. To address this, we develop physics-informed neural networks and Fourier-based artificial intelligence-driven surrogate solvers, which are computationally efficient without compromising accuracy. To actuate intervention policies in a dynamic fashion, we also incorporate a hybrid control mechanism integrating the use of reinforcement learning with classical mathematical optimization to facilitate adaptive policymaking that benefits from data. Unlike existing work, our framework is rigorously evaluated on real-world epidemiological datasets from the World Health Organization and the Centers for Disease Control and Prevention, and tested extensively for out of- the-box adaptability to cybersecurity (cyber malware), social rumor, and financial contagion problems. We also propose a data-free generative model (Fair4Free) that improves fairness, privacy, and utility in synthetic dataset generation, allowing its use even for constrained-data settings. Experimental evidence indicates that our holistic approach enhances the accuracy of predictive performance compared to baselines, with both lower computational cost and cross-domain generalizability to unprecedented settings. Finally, we set a new state-of-the-art for EpiModel by end-to-end training on fair data.

Introduction:A central limitation of existing temporal image analysis and video understanding models lies in their reliance on explicit motion cues, dense supervision, or auxiliary modalities, which constrains their ability to infer latent temporal structure, evolving semantic states, and long-range dependencies from silent image sequences. This limitation becomes critical in settings where temporal meaning emerges implicitly from stable visual representations rather than explicit frame-to-frame dynamics. Methods:In this work, we propose CATS (Context-Aware Temporal Synthesis), a mathematically grounded and interpretable framework for temporal reasoning that operates directly on silent image sequences and general temporal signals. CATS integrates curvature-aware temporal alignment, symmetry-enforced attention, slot-based nonlinear recurrence, and semantic memory fusion to model temporal coherence under noise, partial observability, and unordered inputs. Unlike conventional spatiotemporal architectures, CATS does not assume fixed temporal ordering or handcrafted motion representations, enabling robust temporal abstraction across heterogeneous domains. We validate the proposed framework primarily on silent egocentric video understanding tasks and further assess its robustness and generality through controlled cross-domain temporal stress tests, including stochastic diffusion modeling (ANDI), reinforcement-based temporal alignment, and cyber-physical time-series forecasting. Results and discussion:In particular, we demonstrate that the same architecture trained on visual data transfers effectively to the Anomalous Diffusion (ANDI) benchmark, where CATS organizes particle trajectories in latent time and separates diffusion regimes without architectural modification. This cross-domain consistency confirms that CATS captures intrinsic temporal structure rather than dataset-specific cues. Across visual and non-visual tasks, CATS consistently outperforms competitive baselines, achieving up to 15% relative improvement in mAP and F 1-score on egocentric video understanding, stable regime separation and accuracy gains on anomalous diffusion dynamics, and lower forecasting error in cyber-physical time-series prediction, while maintaining stable convergence under CPU-only constraints and providing interpretable attention and memory dynamics. By unifying temporal alignment, memory, and reasoning within a principled mathematical framework, CATS establishes a domain-agnostic approach to temporal understanding, advancing the state of the art in interpretable temporal reasoning for computer vision and beyond.

The utilization of deep convolutional neural networks for the purpose of diagnosing diseases in the skin area has proven to yield similar accuracy levels to those obtained by dermatologists in different studies. Nevertheless, many challenges are still present, including underperformance and poor generalization in some cases, as well as low interpretability related to the use of black box models. This creates major obstacles for practical implementation since it requires explainability in addition to accurate diagnostics, making it necessary to find a solution. To overcome the mentioned difficulties, an explainable deep learning framework for skin lesion classification (EDLF-SLC) is developed within this study. The framework makes use of various approaches in machine learning and explainable artificial intelligence (XAI) to ensure improvements in terms of both accuracy and interpretability in a three-stage manner. In the initial stage, deep representations extracted from multiple pretrained CNN architectures are fused to preserve complementary discriminative information learned by different network architectures before classification using an SVM with a radial basis function kernel. Next, Support Vector Machine (SVM) classifiers with a radial basis function kernel are used to classify the obtained features. Finally, the predictions made by the model are interpreted through local interpretable model-agnostic explanations (LIME). Experimental results show that EDLF-SLC reaches 88.0% in accuracy, 89.0% in precision, 87.0% in recall, and 88.0% in F1 score, demonstrating competitive performance compared with several recently reported methods under the experimental conditions considered in this study.

In recent years, a range of innovative approaches have been developed to tackle decision-making challenges more effectively. These approaches include concepts such as probability theory, fuzzy set theory, intuitionistic fuzzy sets, vague sets, interval mathematics, and rough set theory. These tools each offer unique approaches to managing uncertainty and imprecision, which are common when working with complex systems. One major issue in software development is failure to meet security requirements. This can be caused by a variety of factors, such as unclear security specifications, evolving threat landscapes, or gaps in communication among stakeholders. To address this persistent problem, this paper introduces a new method designed to quantify security requirements more accurately and streamline the decision-making process for assessing security risks. This approach is based on fuzzy soft set theory, which allows for more flexible and precise handling of uncertain or ambiguous information. By using this method, software engineers and security professionals can gain a clearer understanding of security risks and make better-informed decisions when building secure systems. To demonstrate the practical value of the proposed method, a case study is conducted to evaluate its effectiveness. The results of this case study suggest that security administrators can use this method to implement more targeted and efficient security measures, ensuring that the systems they develop are more resilient to potential threats. Received: 20 July 2025 | Revised: 20 October 2025 | Accepted: 31 October 2025 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data are available from the corresponding author upon reasonable request. Author Contribution Statement Dyala Ibrahim: Conceptualization, Software, Resources, Data curation, Writing – review & editing, Supervision, Project administration. Omar Isam Al Mrayat: Conceptualization, Methodology, Software, Resources, Data curation, Writing – review & editing, Supervision, Project administration. Lial Raja Alzabin: Software, Resources, Data curation, Visualization. Ahmad Alajarmeh: Validation, Visualization. Ahmad Reda Alzighaibi: Validation, Formal analysis, Writing – original draft. Hasan Hashim: Investigation. El-Sayed Atlam: Investigation. Malik Jawarneh: Investigation.

Pneumonia remains a serious global health issue, particularly affecting vulnerable groups such as children and the elderly, where timely and accurate diagnosis is critical for effective treatment. Recent advances in deep learning have significantly enhanced pneumonia detection using chest X-rays, yet many current methods still face challenges with interpretability, efficiency, and clinical applicability. In this work, we proposed a YOLOv11-based deep learning framework designed for real-time pneumonia detection, strengthened by the integration of Grad-CAM for visual interpretability. To further enhance robustness, the framework incorporated preprocessing techniques such as Contrast Limited Adaptive Histogram Equalization (CLAHE) for contrast improvement, region-of-interest extraction, and lung segmentation, ensuring both precise localization and improved focus on clinically relevant features. Evaluation on two publicly available datasets confirmed the effectiveness of the approach. On the COVID-19 Radiography Dataset, the system reached a macro-average accuracy of 98.50%, precision of 98.60%, recall of 97.40%, and F1-score of 97.99%. On the Chest X-ray COVID-19 & Pneumonia dataset, it achieved 98.06% accuracy, with corresponding high precision and recall, yielding an F1-score of 98.06%. The Grad-CAM visualizations consistently highlighted pathologically relevant lung regions, providing radiologists with interpretable and trustworthy predictions. Comparative analysis with other recent approaches demonstrated the superiority of the proposed method in both diagnostic accuracy and transparency. With its combination of real-time processing, strong predictive capability, and explainable outputs, the framework represents a reliable and clinically applicable tool for supporting pneumonia and COVID-19 diagnosis in diverse healthcare settings.

Colorectal cancer (CRC) is one of the most common causes of cancer-related deaths globally, making a timely and reliable diagnosis essential. Manual histopathology assessment, though clinically standard, is prone to observer variability, while existing computational approaches often trade accuracy for interpretability, limiting their clinical utility. This paper introduces a deep learning framework that couples the YOLOv8 architecture for multiclass lesion classification with EigenCAM for transparent model explanations. The pipeline integrates three core stages: (i) acquisition and preprocessing of 5000 hematoxylin-and-eosin-stained slides from the University Medical Center Mannheim, categorized into eight tissue types; (ii) comparative evaluation of five YOLOv8 variants (Nano, Small, Medium, Large, XLarge); and (iii) interpretability through EigenCAM visualizations to highlight discriminative regions driving predictions. Extensive statistical validation (including box plots, empirical cumulative distribution functions, Bland–Altman plots, and pair plots) demonstrated the robustness and reliability of the framework. The YOLOv8 XLarge model achieved 99.38% training accuracy and 96.62% testing accuracy, outperforming recent CNN- and Transformer-based systems (≤95%). This framework establishes a clinically dependable foundation for AI-assisted CRC diagnosis by uniting high precision with visual interpretability. It represents a significant step toward real-world deployment in pathology workflows.

Cyberhate presents a multifaceted, context-sensitive challenge that existing detection methods often struggle to tackle effectively. Large language models (LLMs) exhibit considerable potential for improving cyberhate detection due to their advanced contextual understanding. However, detection alone is insufficient; it is crucial for software to also promote healthier user behaviors and empower individuals to actively confront the spread of cyberhate. This study investigates whether integrating large language models (LLMs) with persuasive technology (PT) can effectively detect cyberhate and encourage prosocial user behavior in digital spaces. Through an empirical study, we examine users’ perceptions of a self-monitoring persuasive strategy designed to reduce cyberhate. Specifically, the study introduces the Comment Analysis Feature to limit cyberhate spread, utilizing a prompt-based fine-tuning approach combined with LLMs. By framing users’ comments within the relevant context of cyberhate, the feature classifies input as either cyberhate or non-cyberhate and generates context-aware alternative statements when necessary to encourage more positive communication. A case study evaluated its real-world performance, examining user comments, detection accuracy, and the impact of alternative statements on user engagement and perception. The findings indicate that while most of the users (83%) found the suggestions clear and helpful, some resisted them, either because they felt the changes were irrelevant or misaligned with their intended expression (15%) or because they perceived them as a form of censorship (36%). However, a substantial number of users (40%) believed the interventions enhanced their language and overall commenting tone, with 68% suggesting they could have a positive long-term impact on reducing cyberhate. These insights highlight the potential of combining LLMs and PT to promote healthier online discourse while underscoring the need to address user concerns regarding relevance, intent, and freedom of expression.

The scheme of water resources management is necessary for reducing water scarcity in arid areas and improving water availability in general. However, water leak detection and irrigation scheduling traditional AI models are often computationally intensive and require complex hyperparameter tuning, making them less scalable. This study presents an artificial intelligence-based optimization framework that improves forecasting accuracy, computational speed, and real-time adaptability. The architecture combines the ensemble-learning algorithms (XGBoost, LightGBM), hybrid AIs (XGBoost + Autoencoder), and metaheuristic feature selection (GA, PSO, SA) for making intelligent decisions. Moreover, ontology-based feature structuring enhances interpretability, while hyperparameter tuning (GridSearchCV, Bayesian Optimization) and model compression techniques (pruning, quantization, knowledge distillation) ensure computational efficiency. A large number of experiments on real-world IoT sensor data testify to the effectiveness of the framework. It achieves 0.992 AUC-ROC scores for leak detection, an RMSE of 0.227 hours for irrigation scheduling, and an overall accuracy of 94.8%. Additional performance measures comprise precision (89.0%), recall (95.2%), F1-score (0.92), and inference speed (0.003 ms/sample). Although quantization has reduced the computational overhead, we still see a 13.02% increase in the model size as seen in Experiment 6, leading to a trade-off that needs to be optimized further. This study offers a deployable AI-based model for sustainable water management by tackling the issues of scalability, computational cost, and limitations in benchmark evaluation. By virtue of the empirical validation and comparative analysis of the framework, it has been shown to perform better than the regular methods, proving that the methodology can act as a step forward in the field of real-time, AI-assisted irrigation and leak detection systems.

The recognition of mental disorder symptoms is crucial for timely management and reduction of recurring symptoms and disabilities. The ability to predict and explain mental health challenges can enable earlier intervention and more effective, individualized care plans, improving the overall well-being of people with autism. Consequently, artificial intelligence (AI) methods have been applied to assist psychologists and psychiatrists in decision-making processes by analyzing patients' medical histories and behavioral data. The current models for diagnosing mental health disorders (MHD) suffer from a lack of interpretability. This study introduces the Explainable Mental Health Disorders (EMHD) model, a robust framework that leverages machine learning algorithms and Explainable Artificial Intelligence (XAI) to identify mental health disorders in young children, including toddlers. The EMHD consists of two main components: an ensemble model and Explainable Artificial Intelligence (XAI). First, an ensemble model known as Voting, which uses different feature selection techniques, namely Mutual Information (Mutinfo), Analysis of Variance (ANOVA) and Recursive Feature Elimination (RFE), is applied to classify the MHD dataset. Second, XAI is integrated into the proposed framework to provide transparency and explanations for the model's decision-making process. To achieve that, the model are explained using a well-known XAI technique called Shapley Additive Explanations (SHAP). The proposed EMHD demonstrates superior performance across all evaluation metrics, achieving an accuracy, precision, recall, and F1-Score of 1.0, in comparison to other baseline models. Furthermore, the study highlights the potential of XAI to provide personalized and actionable insights to mental health professionals who work with autistic individuals. Finally, this study can address the pressing MHD crisis in Saudi Arabia and significantly improve early MHD diagnosis.

Social media platforms serve as significant spaces for users to have conversations, discussions and express their opinions. However, anonymity provided to users on these platforms allows the spread of hate speech and other offensive material. Due to the wide-ranging nature of these platforms, there is a critical need to automatically detect and report occurrences of hate speech. There are various detection methods, but many of them operate as black boxes, lacking interpretability and explainability by design. To address the lack of interpretability, this study explores the development of an interpretable framework to detect hate speech in Arabic using large language models (LLMs). The proposed approach combines advanced natural language processing techniques with interpretable machine learning methods to enhance understanding of model decisions. The experimental results demonstrate that the model achieves high accuracy while maintaining interpretability, enabling users to understand the reasoning behind the detections. The proposed method achieves an accuracy of 0.846%, with a precision of 0.843% and a recall of 0.846%, outperforming existing Arabic hate speech detection models. These results show the effectiveness of combining LLM with interpretability for this critical task, providing a reliable and transparent solution for automated moderation of harmful content.

The rapid proliferation of the Internet of Things (IoT) has increased the risk of sophisticated cyber-attacks, particularly botnets, which can lead to privacy breaches, service disruptions, and infrastructure damage. Traditional security solutions, such as firewalls and signature-based (IDS), are often ineffective due to their static nature and inability to adapt to evolving threats. To address these limitations, this paper proposes Bot-EnsIDS, a dynamic and intelligent intrusion detection system that integrates ensemble bio-inspired optimizer algorithms with a hybrid deep learning classifier. The proposed approach addresses key challenges in existing IDS frameworks, including the inability to efficiently handle high-dimensional IoT traffic data and adapt to detecting newly emerging or obfuscated attacks. The Bot-EnsIDS framework comprises multiple stages, including data preprocessing, a novel multi-objective function for feature selection, and an ensemble-based optimization using Particle Swarm Optimization (PSO) and Gorilla Troops Optimizer (GTO) to extract mutual features. It also incorporates enhanced automatic data augmentation using a modified Generative Adversarial Network (GAN), followed by detection using a hybrid CNN-LSTM deep learning model capable of efficiently detecting botnet attacks by learning from spatial and temporal data features through integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) algorithms. The empirical evaluation of the proposed approach is performed using the BoT-IoT benchmark dataset and demonstrates significant improvements in detection performance over 300 training epochs. The findings revealed that the proposed approach achieves an enhanced accuracy of 97%, recall of 97.5%, precision of 97.5%, and F-measure of 97.5%, indicating significant enhancements in correctly identifying normal and anomalous traffic. The false-positive rate dropped to 0.025, highlighting the system’s precision and reduced false alarms. Overall, this paper introduces a sophisticated IDS architecture that addresses existing gaps and sets a new standard in IoT security by integrating hybrid deep learning techniques and bio-inspired algorithms, promising a more secure IoT ecosystem.

Saudi Arabic Sign Language (SArSL) recognition poses significant challenges due to its complex spatio-temporal structure and the scarcity of annotated datasets. This paper introduces a self-supervised learning framework built upon the Video Momentum Contrast (VideoMoCo) paradigm integrated with a 3D ResNet-50 backbone, designed to jointly capture spatial and temporal gesture dependencies. The proposed model is pretrained on 18,000 unlabeled gesture videos and subsequently fine-tuned on the KARSL-502 dataset containing 15,400 labeled samples covering 502 distinct classes. Experimental evaluation shows that the model attains an F1-score of 92.7%, outperforming CNN-LSTM (86.0%) and Two-Stream CNN (84.5%) baselines-an improvement of nearly 9% points. Beyond accuracy, the framework demonstrates strong robustness to class imbalance, motion variation, and visual noise, while maintaining efficient deployment performance with an inference latency of 12 ms per batch. The ablation study verifies the contribution of the momentum encoder and large negative sample queue in achieving stable and discriminative feature learning. Overall, the VideoMoCo-ResNet-50 framework establishes a scalable and inclusive foundation for real-time SArSL recognition, advancing accessibility for the Saudi Deaf community and supporting future multimodal extensions.

In the poultry food industry, eggshell color is recognized as a crucial quality indicator that influences consumer preference and market value. Traditional classification methods, such as manual inspection and spectrometry, are inefficient and prone to errors. To address this issue, the current study focused on developing an AI-driven system that employs the MobileNetV2 model optimized through a genetic algorithm (GA). Specifically, an attention-enhanced MobileNetV2 architecture was utilized, incorporating 90 frozen layers, dual attention modules, and a dropout rate of 0.254. The results showed a perfect (100%) classification accuracy achieved on a dataset of 2,483 egg images across three classes: brown, platform, and white. The research highlights the significant impact of architectural components such as second dense layers and attention mechanisms on classification performance. This approach offers a robust framework for precise eggshell color classification that can be adapted for quality assessment in poultry production and genetic research applications.

Adaptive technologies have become more sophisticated with Artificial Intelligence (AI) and the Internet of Things (IoT), providing world-changing solutions to help people living with disabilities live better lives. In this article, we discuss the potential of AI and IoT to address issues related to Down Syndrome (DS), Autism Spectrum Disorder (ASD), Mobility Impairment (MI), Hearing Impairment (HI), Attention-Deficit/Hyperactivity Disorder (ADHD), and Visual Impairment (VI). In addition, we propose an analytical framework for evaluating AI and IoT disability assistance prototypes. The framework consists of three different layers: Disability Monitoring, Disability Analysis, and Disability Assistance layers. In each layer, a set of dimensions are identified (e.g., technology, data, security, customization, and response time) and used as criteria to evaluate the research prototypes. Moreover, we evaluate 30 representative AI and IoT disability assistance research prototypes published from 2020 to 2024. The evaluation offers valuable insights into the new strategies, technologies, and approaches that will define AI and IoT disability support in the future. While these technologies have promise in enabling access, autonomy, and interfacing, there remain major open research issues such as data privacy, security, cost, scalability, and real-time response. Furthermore, we discuss future research directions to tackle these issues and allow the people with disabilities community to enhance their quality of life and be more independent.

Children with autism often struggle with expressing and understanding emotions, which can lead to challenges in their social interactions and emotional well-being. In recent years, deep learning techniques and Internet of Things (IoT) technologies have shown promise in addressing these difficulties by providing automated systems for emotion recognition and understanding. However, the lack of interpretability in these systems hinders their adoption in real-world scenarios. This paper presents an Interpretable IoT-based EfficientNet Model (IIENM) for emotion recognition among children with autism. The proposed model aims to address the challenges of accurately identifying and understanding emotions in this specific group by utilizing the capabilities of a deep learning model and integrating IoT technologies. IIENM first utilizes EfficientNet trained on two datasets of facial expressions to accurately classify emotions into several categories, such as sadness, anger, and happiness. Additionally, IoT devices are employed to capture real-time data, including facial expressions, enabling a comprehensive understanding of the child’s emotional state. The system also incorporates Explainable Artificial Intelligence (XAI) techniques such as local interpretable model-agnostic explanations (LIME) and gradient-weighted class activation mapping (Grad-CAM) as interpretation methods to highlight the most influential regions in facial images and physiological signals, providing insights into the decision-making process of the model. The proposed system is tested extensively using two publicly available benchmark datasets of autistic children. The results demonstrate its superior performance in emotion recognition compared to existing methods while maintaining a high level of interpretability. The results also show that the proposed model outperforms state-of-the-art methods in terms of precision, recall, F1 score, and accuracy. The model achieved accuracy scores of 0.92 for Dataset 1 and 0.96 for Dataset 2, respectively.

In recent years, deepfakes (DFs)-realistically manipulated media created using artificial intelligence—have raised significant concerns. As this technology evolves, the urgency for effective detection methods to counter misuse intensifies. Computer science researchers are increasingly focused on stopping the spread of deepfakes (DFs) on social media. However, there has been no comprehensive overview of research in this area. This paper presents a systematic literature map that analyzes research on DF spread on social media from 286 primary studies published between 2018 and June 2024. The studies are categorized by their research type, contribution and focus, revealing a predominant emphasis on detection solutions. Notably, there are significant gaps in evaluating these solutions, using digital interventions to curb dissemination, and managing DF propagation. This literature map will aid researchers, practitioners, and policymakers navigate the rapidly evolving field of DF detection by presenting a structured overview of the available knowledge. The findings of this literature map suggest that DF detection is a multidisciplinary field that requires collaboration between experts in computer vision, machine learning, cybersecurity, and media forensics to address its current and future challenges

Network security is a critical issue in the digital era, as the complexity and intelligence of cyber threats pose significant risks to people as well as companies. Traditional anomaly detection techniques often struggle to effectively identify novel and evasive network attacks, prompting the need for more robust and adaptive approaches. This study presents a robust framework for enhancing network security through the integration of ensemble learning and explainable artificial intelligence (XAI) techniques. The proposed approach utilizes the combined strengths of ensemble learning models to enhance the performance and reliability of network anomaly detection. The proposed model achieves high scores across all key metrics: precision (99.99%), accuracy (99.99%), recall (100%), and F1-score (99.99%). Furthermore, the XAI techniques provide valuable insights into the fundamental elements influencing the detection decisions.