Dynamic Multi-Scale Graph-Attention Temporal Network and Adaptive Hybrid Feature Selection Framework for a Blockchain-Integrated Intrusion Detection System in IIoT

This study proposes a scalable and reliable intrusion detection framework for Industrial Internet of Things (IIoT) systems to address challenges such as real-time adaptability, feature relevance, and data integrity. The design combines an adaptive hybrid feature selection mechanism to find the best data characteristics with a dynamic multi-scale graph attention temporal network for spatial-temporal threat detection. To guarantee decentralization and data integrity, blockchain technology is used. Metrics such as accuracy, precision, recall, and latency were used to assess the system. It achieved 96.2% detection accuracy, 95.5% precision, and a latency of just 17.8 milliseconds, which was much better than previous systems. The framework provides scalable and dependable cybersecurity for IIoT systems by fusing blockchain, AI-driven graph attention methods, and hybrid feature selection. This allows for real-time, safe, and effective monitoring in Industry 5.0 settings.