An Explainable Ensemble Deep Learning Approach for Intrusion Detection in Industrial Internet of Things
Idris Yusuf Safana
Obunadike G.N.
Yusuf Surajo
Abstract
The Industrial Internet of Things (IIoT) has become vital to the operation of critical infrastructures; however, its widespread adoption is hindered by security vulnerabilities that expose IIoT systems to increasingly sophisticated cyberattacks. Existing intrusion detection systems (IDSs) for IIoT primarily focus on detection accuracy while offering limited interpretability, thereby reducing their practical trustworthiness and deployment in real-world industrial environments This research presents an explainable ensemble deep learning-based intrusion detection system (IDS) for IIoT networks, combining Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) units, and Autoencoders. The proposed ensemble integrates spatial feature extraction, temporal dependency modeling, and reconstruction-based learning through a unified training and decision-fusion mechanism The research is aimed at resolving the two-fold challenges of detecting more accurate results and giving clear interpretation of the model, which is vital for cybersecurity in real-world applications. This research employed the ToN-IoT dataset comprising of various attack scenarios in IIoT environments, the introduced system showed better performance in both binary and multi-class intrusion detection tasks. The binary model classification showed an accuracy of 98.5%, precision of 98.2%, recall of 97.9%, F1-score of 98.0%, and an AUC-ROC value of 0.992. In multi-class classification model, the system achieved an accuracy of 94.2%, precision of 93.5%, recall of 94.1%, F1-score of 93.8%, and an AUC-ROC value of 0.987. Additionally, the incorporation of Explainable AI (XAI) techniques such as Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) provided transparency in the decision-making process. Unlike existing approaches, this study uniquely combines ensemble deep learning with post-hoc explainability to simultaneously achieve high detection accuracy and interpretable intrusion detection for IIoT systems.
References