We present SECURE-XED, a unified, explainable, and adversarial-aware learning system for two traditionally disjoint domains: Android malware classification and deepfake detection. The system uses a shared convolutional neural network backbone with domain-specific input adaptation and two task-specific heads. It integrates SHapley Additive exPlanations for explanation, the Fast Gradient Sign Method and Projected Gradient Descent for adversarial evaluation under Structural Similarity Index Measure guardrails, and a selective-activation controller that enables behavioral malware features when confidence is low or evasion is suspected. A hybrid Seagull Optimization Algorithm–Imperialist Competitive Algorithm procedure jointly tunes clean accuracy, robustness under projected-gradient attacks, expected calibration error, parameter count, and inference latency. SECURE-XED is evaluated on standard Android malware corpora and the FaceForensics++, Celeb-DF, and Deepfake Detection Challenge benchmarks. Under Projected Gradient Descent, Android malware accuracy decreases by approximately 19–21% with a perturbation budget of 0.1 over 40 steps, while deepfake accuracy decreases by 16.6–17.6% with a perturbation budget of 0.03 over 10 steps, demonstrating domain-dependent adversarial sensitivity. Under clean deepfake inference conditions, excluding post-prediction explanation generation and adversarial processing, the unified configuration reduces average inference latency from 34.1 to 26.9 milliseconds per frame and approximately halves the parameter requirement relative to maintaining separate model instances, while retaining equal or slightly better clean accuracy. A classroom-oriented simulator exposes explanation overlays and adversarial controls on real frames. A six-participant human-grounded pilot descriptively showed a 22-percentage-point change in Region Identification Accuracy, a 12-second reduction in decision time, and a 0.9-point Likert change in perceived clarity and trust; these pilot outcomes are descriptive and not inferential. The resulting framework combines shared cross-domain representation, multi-objective optimization, adversarial evaluation, explainability, and model-in-the-loop educational interaction within a single modular architecture.

The increasing digitalization of smart grid systems has introduced new cybersecurity challenges, particularly at the supervisory control and data acquisition (SCADA) edge gateways where resource constraints, latency sensitivity, and privacy concerns limit the applicability of centralized security solutions. This paper presents FED-SCADA, a novel federated intrusion detection system (IDS) that integrates Spiking Neural Networks (SNNs) for energy-efficient inference and Differential Evolution (DE) for optimizing model convergence in decentralized, non-independent and identically distributed (non-IID) environments. The proposed architecture enables real-time, privacy-preserving intrusion detection across distributed SCADA subsystems in a smart grid context. FED-SCADA is evaluated using three public IIoT/SCADA datasets: TON_IoT, Edge-IIoTset, and SWaT. FED-SCADA achieves a detection accuracy of 96.4%, inference latency of 28 ms, and energy consumption of 1.1 mJ per sample, demonstrating strong performance under real-time and energy-constrained conditions outperforming base-line federated learning models such as FedAvg-CNN and FedSVM. A detailed methodology flowchart and pseudocode are included to support reproducibility. To the best of our knowledge, this is the first study to combine neuromorphic computing, evolutionary optimization, and federated learning for trustworthy and efficient smart grid cybersecurity.

The increasing complexity and dynamism of modern cyber threats necessitate intelligent and adaptive network intrusion detection systems (NIDS). This paper proposes a novel hybrid metaheuristic approach that combines the Lion Optimization Algorithm (LOA) with the Virus Colony Search (VCS), enhanced by adaptive parameter tuning mechanisms. The proposed LOA-VCS hybrid algorithm addresses limitations in prior single and hybrid metaheuristic by alternating exploration and exploitation strategies across epochs, optimizing detection performance in high-dimensional feature spaces. Unlike previous hybrid metaheuristics that use fixed or non-adaptive control, our model uniquely alternates LOA and VCS phases adaptively across epochs to enhance convergence and detection robustness. A real-world intrusion detection dataset evaluated the LOA-VCS model with 98.4% detection accuracy, an F1-score of 0.976, and an AUC of 0.986, consistently outperforming the standalone LOA and VCS baselines. These results emphasize the power of adaptive hybrid met heuristics in maintaining low false alarms while ensuring strong recall for NIDS. The proposed approach can be deployed in scalable, high-speed systems in today’s contemporary cyber security environments.