Adaptive Weighted Ensemble Learning for Intrusion Detection in Industrial IoT and Edge Computing

The rapid adoption of Industry 4.0/5.0 and Industrial Internet of Things (IIoT) has introduced new cybersecurity vulnerabilities in industrial operations. IIoT devices, despite their widespread deployment, face resource constraints that make traditional security solutions ineffective. While Machine Learning (ML)-based Intrusion Detection Systems (IDS) have emerged as promising solutions, they often face limitations when dealing with industrial-scale data sets and multi-class attack detection scenarios. This paper proposes an enhanced weighted ensemble learning approach that overcomes these limitations while maintaining robust intrusion detection capabilities in IIoT and Edge Computing environments. Our solution features a weighted method that processes data into strategic subsets, creating an optimized ensemble of ML models. The model integrates data collection, preprocessing optimization, model training, and evaluation phases. Our approach, validated on the Edge-IIoTset public dataset, yielded impressive results: the TabPFN model achieved a $\mathbf{9 9. 8 1 \%}$ F1-Score, and all other implemented models demonstrated significant performance improvements over their respective baseline versions.