Internet of Things (IoT) and Internet of Medical Things (IoMT) networks provide a massive amount of data. These types of data need a protection level, such as an intrusion detection framework. Deep learning models become a powerful tool for this purpose. Therefore, this work proposes an intrusion detection framework based on a deep learning technique which employs TabTransformer and self-attention mechanisms to imprison intricate dependencies among tabular features and detect abnormal attack behaviors. Precisely, each numerical feature is mapped into a learnable embedding vector and augmented with positional encodings to preserve feature identity and inter-feature relationships within the embedding space. The main task for the proposed model is to achieve binary classification tasks the model should classify the traffic data as either normal or abnormal. Furthermore, the model utilized a benchmark dataset such as the CICIoMT2024. Furthermore, this type of dataset faces issues, such as imbalance. So, the system integrates SMOTE-based data balancing, Stratified K-Fold Cross-Validation, and threshold optimization to ensure fairness and reproducibility to accomplish a binary classification task. As a consequence, experiments on the CICIoMT2024 dataset yield superior results, achieving a mean accuracy of 99.85. Through SHAP-based interpretability, key features influencing model predictions are identified, confirming the framework’s transparency, robustness, and suitability for real-world ARP intrusion detection.

Cyber-attacks become manifested as a series of behavioral patterns, but not as an event, and many current security regimes stay based upon a static detection and fixed trust implementation. Such incongruence restricts their capability to act in a dependable manner in fluctuating and unpredictable threat situations. The existing artificial intelligence-based cybersecurity products mainly focus on the detection performance. Due to this, such systems will still be vulnerable to false positives, erratic reactions, and degradation of performance over time. This paper proposes a behavior-sensitive cybersecurity model that brings together sequential behavioral modelling, risk-adaptive cryptography implementation, and integrity-guaranteed learning in an architecture with closed loops. The temporally structured patterns of activity are considered as behavioral risk, which allows making proportional, not binary, trust decisions. Cryptographic policies are adaptively changed based on the inferenced risk, whereas learning updates are explicitly secured to maintain the model reliability as time goes by. The experimental findings indicate that the proposed framework can obtain a detection accuracy of 96.7% and F 1-score of 96.0, as well as a false positive rate decreased to 3.1%. Moreover, the adaptive response latency is also decreased by a factor of about 20-30% relative to the representative baselines and also enhanced stability in response to adversarial noise. These results indicate behavior-based intelligence.

The recent explosion of cyber threats in big data ecosystems has exposed the vulnerability of black-box machine learning models that prioritize accuracy over explainability. Traditional cybersecurity mechanisms do not create a sense as to why a prediction is provided and the analysts are left with doubts and the response mechanisms are slow. This is not very transparent and therefore compromises the operational trust and the traceability of high-risk decisions in real-time defense infrastructures. Current explainable AI (XAI) methods, despite their usefulness, are mostly stagnant, not connected to the changing situation of network behavior and human monitoring. They seldom inculcate the feedback mechanisms that can adjust the model explanations with new threat patterns. The research paper introduces an Explainable Artificial Intelligence and Data-Driven Security Analytics Framework that is hybrid and serves to combine global interpretability with SHAP, local reasoning with LIME, and adaptive refinement with an analyst-in-loop feedback layer. The architecture converts the threat detection to a self-fixing transparent process in which the analytical reasoning is dynamically developed with the data flow. In experimental tests of less than 10k concurrent traffic conditions, experimental results indicate that the system had a detection accuracy of 98.4 and Spearman correlation (0.91) between predicted risk scores and real levels of severity with a quantifiable 3.9% stability improvement over the similar XAI-based systems of intrusion detection. The combination of explainability with real-time analytics will further enhance the accuracy of detecting cyber threats and its interpretability reliability as well. The findings indicate the essential change to credible, self-explanatory, and adaptive security intelligence of the next-generation data networks.

The exponential growth of Internet of Things (IoT) devices integrated with fifth-generation (5G) wireless networks has created unprecedented opportunities for ultra-low-latency applications while introducing complex security vulnerabilities and computational challenges. This paper presents a comprehensive framework for deploying adaptive lightweight Convolutional Neural Networks (CNNs) in 5G-enabled IoT environments to address intrusion detection, intelligent traffic classification, and dynamic resource optimization. We propose a novel multi-objective optimization approach that integrates Adaptive Depthwise Separable Convolutions (ADSC), Dynamic Quantization-Aware Training (DQAT), and Real time Pruning Strategy (RPS) specifically designed for 5G network slicing architectures. Our methodology incorporates federated learning principles, edge-cloud collaboration, and context-aware adaptation mechanisms. Comprehensive evaluation on multiple datasets, including NF-ToN-IoT-v2, NSL-KDD, and CICIDS-2017, demonstrates superior performance with 97.8% accuracy in multi-class attack detection, 76% reduction in computational overhead, 71% decrease in energy consumption, and 42% improvement in network throughput. The framework achieves inference times under 8.5ms on edge devices while maintaining robust security postures across heterogeneous IoT deployments. Statistical significance testing and large-scale ablation studies verify the effectiveness of each of the suggested elements.

Current network security technologies face new threats from determined attackers employing advanced evasion techniques such as IP spoofing, tiny fragment attacks, tunneling, and HTML smuggling. Conventional intrusion detection and prevention systems (IDS/IPS) have significant limitations in detecting zero-day attacks and sophisticated threats that can continuously alter their attack vectors. This paper presents a novel deep learning-driven, multilayer intrusion detection and prevention framework that integrates network-based IDS/IPS, host-based intrusion detection systems (HIDS), and honeypot technologies with advanced machine learning models, including graph neural networks (GNNs), autoencoders, and transformers. The framework employs adaptive learning mechanisms to enhance resilience against evasion techniques while maintaining low false positive rates. Experimental evaluation using diverse attack datasets demonstrates superior performance, achieving 97.3% detection accuracy for zero-day attacks and 94.8% resilience against advanced evasion techniques, significantly outperforming existing state-of-the-art solutions. The proposed framework contributes to cybersecurity research by introducing innovative multilayer correlation mechanisms, adaptive threat modeling, and evasion-resilient detection algorithms.

Autonomous ground robots are increasingly relied upon in dynamic environments where reliable navigation and context-aware interaction are essential. However, existing robotic control systems often rely on cloud-based reinforcement learning (RL) frameworks or static algorithms that fail to adapt in real-time to noisy, unpredictable scenarios. These models typically overlook the constraints of edge deployment and lack robust integration with human interaction modalities such as voice and semantic object awareness. To address these limitations, this work proposes a fully embedded, IoT-enabled ground robot powered by a reinforcement learning-based adaptive control framework. The system leverages Raspberry Pi 4B+ as its core computational unit, integrating MQTT-driven communication, multimodal interaction through speech and vision, and lightweight policy convergence for obstacle-aware navigation. A novel RL-based state-action pipeline is trained and deployed entirely on-device, ensuring real-time responsiveness without external computation. Experimental evaluations show that the proposed framework reduces navigation errors by 22% and improves interaction latency by 37% over traditional PID and A*-based systems. The RL model converges in under 2200 episodes, with stable reward curves and high reliability across variable acoustic and physical terrains. This study showcases how low-cost, edge-based robots can achieve high autonomy and situational awareness contributing to future advancements in resilient, self-adaptive robotic systems within smart and resource-constrained environments.