The evolving threat landscape demands intrusion detection systems that adapt quickly to novel attack patterns and operate across heterogeneous environments. Recent studies show that Generative Adversarial Networks (GANs) can improve intrusion detection performance by generating synthetic attack traffic, balancing imbalanced datasets, enhancing adversarial robustness, and serving as anomaly detectors. This survey provides a comprehensive and systematic review of GAN-based intrusion detection system (IDS) research, analyzing the architectures employed—including Wasserstein GANs, conditional GANs, self-attention GANs, and specialized multi-generator designs—together with their applications, datasets, and evaluation metrics. Unlike previous surveys, we extend the scope to resource-constrained Internet of Things (IoT) and federated scenarios, where lightweight and tabular GANs can process sensor data and operate on edge devices. We also examine deployments in software-defined networking environments. We propose a unified evaluation framework that reports class-wise precision, recall and macro-F1-scores, per-attack metrics, computational cost, and statistical similarity tests, and we emphasize the need for interpretable and multi-modal approaches that fuse network flows with logs or threat intelligence. Emerging paradigms including GANs combined with large language models, quantum GANs, diffusion models, and reinforcement learning are surveyed, and open challenges such as training instability, mode collapse, hyper-parameter tuning, and ethical dual-use concerns are discussed. By synthesizing recent advances and outlining future research directions, this survey provides a comprehensive and forward-looking reference for practitioners and researchers developing robust, privacy-preserving, and adaptive GAN-based intrusion detection systems.

Sustainable AI merges artificial intelligence (AI) development with ecological, social, and economic resilience. It addresses energy consumption, resource management, ethical implications, and policy integration. This chapter explores current practices, identifies challenges, and uncovers opportunities for developing systems that align with sustainable goals. It reviews existing research on carbon footprints, data governance, ethical frameworks, public policy, and cross-sector collaboration. It also analyzes potential methods to measure and mitigate AI's environmental impact, with illustrative tables and diagrams spread throughout. Finally, the chapter offers an outlook on integrating these findings into a cohesive vision for sustainable AI, bridging academic and industry efforts, and enhancing practical governance frameworks.

Colorectal Cancer (CRC) generally arises from polyps in the colon or rectum. The survival rates and treatments get a great boost when the cancer is detected early. The initial stages of cancers may not show any symptoms, and thus the diagnosis often gets delayed. This study proposes a ML framework called Efficient ML Framework for Early CRC Prediction Using Efficient Net and Vision Transformers (EPCCEV) for early and accurate prediction of CRC using medication data from a Danish registry of the years 2020-2023 to predict the outcomes for 2024-2025. It divides the medication histories into various time intervals to get the idea of drug exposure in the short and long terms. Major steps of the work include data cleaning, class imbalance problem treatment, combination of ensemble Machine Learning (ML) and Deep Learning (DL) for feature extraction and model development. EfficientNetV2 is used to catch the nonlinear relationships, and the Vision Transformer (ViT) is deployed for capturing the contextual dependencies, while hyper parameter tuning and regularization are used for optimization.

Stroke prediction and prevention is an important focus in healthcare due to the significant morbidity and mortality associated with strokes. In this study, we investigate using Generative Adversarial Networks (GANs) to augment a stroke dataset and evaluate the effects on prediction performance. The original dataset contained patient medical records and demographics used to predict stroke occurrence. We trained a GAN on these data and generated synthetic samples to augment the training set. Five machine learning models were developed on the original and augmented datasets, including decision tree, k-nearest neighbors, random forest, Support Vector Machine (SVM), and logistic regression classifiers. Experiments indicate statistically significant improvements in prediction accuracy, F1 score, specificity, and sensitivity with GAN augmentation across all models. The random forest classifier achieved the highest average accuracy of 0.981 on augmented data, versus 0.967 on original data. GANs prove effective for tackling class imbalance and enabling more robust stroke prediction from limited real-world data. This demonstrates the potential of data augmentation and generative models to enhance healthcare Artificial Intelligence (AI) applications.

Phishing detection is challenging when enterprise messages mix dialectal Arabic and English, include irregular spelling, and must be processed under strict data-sovereignty policies. We present a bilingual detection framework that adapts an open-weight transformer using Low-Rank Adaptation (LoRA) while keeping base parameters frozen. The pipeline combines policy-driven redaction, email-aware normalization, joint subword tokenization tuned for code-switching, and signed adapter artifacts for controlled updates. We outline an experimental protocol using verified public email corpora and a parallel English–Arabic phishing dataset, and we specify a deployment architecture for on-premise or sovereign cloud inference with audit logging and key management. The framework targets reliable triage without exporting sensitive content.

This study evaluates eight machine-learning regression models for estimating serum vitamin D level as a support tool for vitamin D deficiency assessment. A cohort dataset of 100 individuals (dataset 132) was analyzed using Support Vector Machine (SVM), Random Forest (RF), Artificial Neural Network (ANN/MLP), Linear Regression (LR), Elastic Net (EN), Ridge Regression (RR), Lasso Regression (LSR), and RANSAC Regressor (RAN). Model performance was assessed over 30 repeated runs using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination (R2). SVM yielded the strongest overall results (MAE = 1.841, MSE = 32.502, mean R2 = 0.9981), followed by RF (MAE = 7.571, MSE = 197.832, mean R2 = 0.9908). ANN showed intermediate performance (mean R2 = 0.8538), whereas RR was the weakest model (mean R2 = 0.4945). To address interpretability, the revised manuscript adds an explainability-oriented feature-attribute analysis derived from the correlation structure of the cohort. The strongest associations with vitamin D were gender (r = 0.64), hemoglobin (r = 0.47), age (r = 0.38), marital status (r = -0.35), and triglycerides (r = 0.32). These findings show that model choice substantially affects predictive performance and that nonlinear models, particularly SVM and RF, can support cost-conscious screening strategies for vitamin D deficiency assessment. Future work should validate the models on larger external cohorts and extend interpretability with model-specific explainability techniques.

Responsible and green artificial intelligence (AI) has become a strategic concern for tourism marketing because the sector increasingly depends on data-intensive personalization, algorithmic targeting, automated service interfaces, and generative content systems. Yet the benefits of AI for destination promotion, dynamic recommendation, revenue optimization, and traveler engagement are inseparable from ethical, legal, and environmental costs. Tourism marketers collect mobility, preference, and behavioral data; they deploy predictive systems that can reproduce bias or erode autonomy; and they increasingly rely on computational infrastructures whose energy use and carbon emissions remain largely invisible. This chapter develops an integrated framework for responsible and green AI in tourism marketing by linking data ethics, privacy governance, and the carbon footprint of intelligence. Drawing on scholarship in tourism, marketing, AI ethics, and Green AI, it offers conceptual grounding, managerial guidance, seven tables, and five figures to support implementation and future research.

Edge intelligence moves learning and inference closer to sensors and users, reducing round-trip latency and limiting raw data transfer. The same shift exposes a mismatch: modern deep networks are accurate but often exceed the memory, compute throughput, and energy budgets of IoT nodes, mobile devices, and embedded accelerators. Lightweight architectures and compression methods are widely available, yet they are frequently applied in sequence-first select a mobile-friendly backbone, then compress until the model fits. This workflow can waste accuracy because architecture choices constrain which compression operators remain stable, and compression changes operator shapes that determine runtime speed and peak activation memory. This paper proposes a budget-aware co-design framework that couples architecture scaling (width, depth, and input resolution) with structured pruning, quantization, and knowledge distillation in a single constrained optimization procedure. A device-aware cost model estimates weight storage, peak activation memory, and latency using a simplified compute-memory balance bound, and a deployment verification loop is described to filter candidates. Using a transparent, illustrative memory-bound analysis across representative edge profiles, we show how activation constraints can dominate in SRAM-limited settings and why early-layer structured pruning can complement INT8 inference by reducing both arithmetic and activation footprints. The paper concludes with design guidelines that connect system-level constraints to concrete architecture and compression choices.

In the modern world of rapid change of the enterprise environment, the system of governance should be dynamic, multi-purpose, and explanatory. To address these requirements, this research paper introduces an AI-Embedded Hybrid Multi-Objective Governance System which allows the adaptive decision-making process within distributed business environment. The framework builds a semantic knowledge graph by combining both structured and unstructured data of ERP, CRM, HRMS, financial, and logistics systems to build an enterprise context. It continuously measures and streamlines five areas of governance compliance, performance, risk, sustainability, and cyber-resilience with a hybrid AI engine consisting of rule-based reasoning, Gradient Boosting models, and Transformer-based deep learning. Reinforcement learning (RL) agents are incorporated to be able to change the policy thresholds in a real-time fashion. Federated Learning also guarantees the privacy-preserving model training units without exchanging raw data. SHAP and causal inference explainability is incorporated to increase trust and auditability. The system was found to have reached up to 92.6 % accuracy in forecasting SLA adherence and 89.5 % in predicting risk index, and increased performance by 15.1 % and reduced risk exposure by 10.1 % on average. Future innovative contributions are to add our unstructured streams of data such as IoT sensors and news feeds, agent-based game theory to negotiate, and blockchain audit trails. The proposed system provides a base towards the real self-governing, intelligent enterprise governance.

Metaverse environments now blend embodied interaction, synthetic media, virtual asset trading, and platform governance in persistent multi-user worlds. But conventional cybersecurity? It's necessary yet insufficient--treating threats as isolated technical incidents ignores their social impact. This chapter proposes an ontology-guided reinforcement learning framework for autonomous threat mitigation. It models actors, assets, interactions, threats, governance constraints, and control options as a linked semantic system that shapes state construction, action boundaries, reward design, explainability, and cross-platform learning. We identify eight knowledge mechanisms that boost context awareness, ensure proportional responses, and enable auditability. Privacy, fairness, scalability, accountability, and adversarial adaptation remain key challenges. Ultimately, trustworthy metaverse defense demands semantically grounded, governance-aware adaptive agents.

This paper introduces DualConnect, an innovative mobility management protocol for 5G networks that utilizes dual connectivity to facilitate seamless handovers. A make-before-break handover execution technique is suggested in which the UE keeps connections to both the source and target base stations during the handover. This stops any loss of packets or service interruptions. Improvements have been made to measurement reporting, target cell selection, conditional handover, and coordination between base stations to make the dual connectivity model work better. An analytical framework determines the ideal duplication time to reduce packet loss. Extensive simulations show that handover interruption time is over 95

This chapter develops an architecture for combining cognitive digital twins with blockchain ecosystems to deliver predictive business intelligence in the industrial metaverse. It argues that digital-twin initiatives create value only when operational telemetry, semantic context, decision models, and human governance are connected through trusted data-sharing and executable coordination mechanisms. Drawing on research in digital twins, industrial metaverse design, blockchain-enabled provenance, smart contracts, human-centric manufacturing, and explainable AI, the chapter proposes a layered reference architecture, performance metrics, implementation logic, and governance controls. It also examines persistent challenges, including semantic fragmentation, model drift, latency, privacy, interoperability, energy cost, and the oracle problem. The chapter concludes with design recommendations and a research agenda for resilient, trustworthy, and economically useful industrial intelligence systems across networked industrial-metaverse platforms and ecosystems.

Vehicular ad hoc networks need rapid authentication under short contacts, intermittent backhaul, and roaming across operators. Kerberos provides efficient ticket-based session keys, but it assumes reachable domain KDCs and offers limited cross-operator audit and revocation. KBXAuth is a hybrid design that keeps Kerberos as the fast admission path and uses a permissioned blockchain only for cross-domain consistency. The ledger stores compact attestations as constant-size commitments with counters and validity windows; it never stores tickets, session keys, or clear vehicle identities. RSUs validate Kerberos tickets locally and consult the ledger only when roaming or risk policy requires it. We specify the protocol workflow, RSU decision logic, and a minimal smart contract for registration, verification, and revocation. We also outline an evaluation plan and practical feasibility considerations.

Software-Defined Networks (SDN) enable rapid, controller-driven mitigation, yet their centralized control plane also creates high-impact failure modes under adversarial traffic and control-plane stress. This paper proposes a MuZero-style, model-based decision layer for autonomous cyber defense in SDN that selects among pre-authorized mitigations (drop, rate-limit, reroute, quarantine, and decoy diversion) under partial observability. The agent plans in a telemetry-derived latent state rather than requiring an explicit topology map, learning short-horizon dynamics that predict decision-relevant security risk, QoS impact, and operational cost. To make Monte Carlo Tree Search practical at flow scale, we introduce a flow-group action abstraction, progressive widening for large action spaces, and a receding-horizon loop aligned with controller timing and rule-update limits. We also outline an evaluation protocol in emulated SDN settings to assess safety, responsiveness, and QoS preservation under attack.

Critical infrastructure incident response increasingly relies on automation to contain cyber-physical threats under time pressure. Reinforcement learning (RL) is attractive for response selection because it can adapt to evolving attack sequences and can mitigate partial observability when the state is augmented with short telemetry and alert history (or when recurrent value networks are used), yet RL policies are typically opaque and therefore difficult to justify in safety-critical operations. This paper develops an explainable reinforcement learning (XRL) framework for automated incident response in supervisory control and data acquisition (SCADA) environments. The framework couples a Deep Q-Network (DQN) response agent with (i) an explicit safety shield that enforces operational constraints and (ii) a post-hoc Explainer Module that produces feature-attribution explanations using Shapley Additive Explanations (SHAP). We specify how SHAP attributions are extracted from the DQN decision process by mapping the selected action’s Q-value to a scalar explanation target, using background states drawn from a governed replay buffer or historian, and applying a latency-bounded gradient-based SHAP approximation (e.g., DeepSHAP/GradientSHAP implementations). We further propose a SCADA control-room workflow in which the agent remains advisory by default, supports graded autonomy for low-risk actions, and produces audit-ready explanation artifacts that align with operator mental models and regulatory expectations. We describe an experimental setup grounded in established industrial datasets (SWaT/WADI and hardware-in-the-loop water distribution data) and testbed practices, and we provide an evaluation plan and indicative findings emphasizing response timeliness, unsafe actuation avoidance, and operator override behavior. The contribution is a precise architecture and operator-centered workflow for deploying XRL in critical infrastructure without relying on undocumented assumptions about autonomy.

Immersive multi-user metaverses combine persistent identity, real-time interaction, virtual assets, and digital-physical linkages, which creates a cybersecurity problem that extends beyond isolated authentication or perimeter controls. This chapter examines how swarm intelligence can support resilient cybersecurity in these environments through distributed sensing, cooperative risk analysis, graded containment, recovery orchestration, and policy-aware federation. To this end, it introduces the Swarm-Intelligence for Resilient Metaverse Cybersecurity (SIRMC) framework as a layered reference architecture spanning avatars, spaces, services, assets, and cybersecurity digital twins under zero-trust principles. In addition, the chapter synthesizes current literature, defines a metaverse-specific threat model, outlines resilience requirements, and provides figures, tables, scenarios, and an implementation blueprint. Overall, the chapter positions resilient metaverse defense as a coordinated capability grounded in local intelligence, twin-assisted foresight, and privacy-aware governance.

Smart buildings rely on dense Internet of Things (IoT) deployments for HVAC, access control, lighting, and occupant services. Connectivity improves efficiency but enlarges the attack surface, enabling adversaries to pivot from commodity devices toward building automation systems. This paper proposes BIMSHIELD, a BIM-contextual digital-twin security architecture that fuses (i) IFC-derived context graphs, (ii) IoT telemetry and network-flow observability, (iii) edge-deployable machine learning for anomaly and intrusion detection, and (iv) Zero-Trust enforcement for micro-segmentation and continuous authorization. BIM semantics are transformed into a BIM Security Context Graph that binds device identity and policy to spaces and subsystem roles, producing facility-readable alerts and auditable responses. We define a trace-driven emulation protocol using a large-scale CIC IoT benchmark (105 devices, 33 attacks, seven categories) to exercise detection-to-enforcement timing under configurable gateway and network budgets.

Cloud platforms increasingly host mixtures of latency-sensitive microservices and data-intensive analytics whose workloads vary at time scales comparable to autoscaler actuation delays. Reactive scaling policies therefore operate with inherent lag, leading to transient service-level objective (SLO) violations during demand surges and unnecessary overprovisioning when demand drops. Predictive autoscaling can reduce this lag, but many deployed and published approaches either (i) rely on point forecasts integrated through heuristic rules, or (ii) solve cost minimization without explicit treatment of forecast uncertainty and SLO risk. This paper presents Predictive Big-Data Model Resource Allocation (PBDMRA), a cloud-native control stack that couples large-scale telemetry processing with probabilistic forecasting, lightweight uncertainty calibration, and risk-bounded allocation. PBDMRA ingests metrics and traces (and can incorporate structured logs when available), constructs features using stream processing, produces multi-horizon predictive quantiles, calibrates empirical coverage via rolling residual adjustments, and computes scaling actions by solving a constrained optimization problem that trades cost against SLO risk and scaling churn. Trace-driven replay across microservice, streaming, and batch workloads indicates that PBDMRA reduces SLO violation minutes by 33-55relative to Kubernetes HPA and a point-forecast baseline, while also reducing scaling actions.

A Threat Detection Model employs analytics, machine learning, or rule-based methods to track system behavior live. This work present the new Enhanced Task Execution and Energy-efficient Quantum Integrated (ETEEQI) model is a technology that is specifically built for the cloud-assisted Internet of Things (IoT) scenario, with the main aim of addressing issues related to security and computational efficiency. The process begins with IoT devices, which apply Compressive Sensing (CS) to achieve a significant data volume reduction for a low-energy transmission that is more bandwidth efficient. ETEEQI is an acronym for a method that achieves the best possible use of the task execution resources across IoT, fog, and cloud environments by using models to assess the communication capacities and processing latencies. Energy consumption is tracked through the use of a Directed Acyclic Graph (DAG) for an energy model to guarantee energy-efficient operations. The simulation analysis is accuracy, communication cost, resource utilization rate, delay, discarded tasks calculation.

The Web of Things (WoT) interconnects a rapidly growing population of smart devices and sensors, enabling innovative applications while exposing an ever‑expanding attack surface. Reinforcement learning (RL) can adaptively detect and mitigate such attacks, yet conventional RL struggles to converge in WoT’s high‑dimensional state‑action spaces. We address this limitation by augmenting RL with the Harris Hawks Optimization (HHO) algorithm. HHO is a recent meta‑heuristic optimization method that balances global exploration with local exploitation and is well suited to large search spaces. We propose an HHO‑based meta‑learning framework that aims to identify hyper‑parameters and network architecture for a deep‑Q network (DQN) defender, maximizing average episodic reward in simulated WoT environments. Experiments on the CIC‑IoT‑2023 and Bot‑IoT datasets show that an HHO‑optimized DQN converges faster and achieves higher accuracy than all tested baselines—including vanilla, double and dueling DQNs, PPO, A3C and Transformer-based agents—illustrating the promise of bio-inspired optimization for adaptive WoT security.