An Explainable Autoencoder-Based Self-Attention Deep Neural Network Framework for Multi-Class IoT Intrusion Detection (Published)
The rapid expansion of Internet of Things (IoT)-enabled electronic systems has introduced significant cybersecurity challenges due to the increasing number of interconnected sensors, embedded devices, and industrial controllers. Conventional intrusion detection systems (IDSs) often suffer from limited adaptability to complex attack patterns, while many deep learning-based approaches provide limited interpretability, restricting their application in security-critical IoT environments. This paper proposes an explainable Autoencoder–Self Attention Deep Neural Network (AE-SA-DNN) framework for intelligent multi-class intrusion detection in IoT networks. The proposed framework integrates autoencoder-based latent feature representation learning, self-attention-driven feature dependency modelling, and SHAP-based explainable artificial intelligence (XAI) analysis to achieve accurate and transparent intrusion detection. The TON_IoT network dataset, consisting of 211,043 traffic samples, 44 network features, and ten intrusion categories, is utilized for experimental evaluation. The framework performs feature preprocessing, dimensionality reduction through autoencoder learning, attention-based classification, and interpretable prediction analysis. Experimental results demonstrate that the proposed AE-SA-DNN achieves superior classification performance compared with conventional machine learning and deep learning approaches, achieving an accuracy of 97.52%, F1-score of 97.55%, and ROC-AUC of 0.975. Furthermore, SHAP analysis identifies the most influential network features contributing to intrusion decisions, improving model transparency and trustworthiness. The proposed framework provides an effective and explainable solution for securing IoT-enabled electronic systems and industrial IoT environments.
Keywords: Explainable Artificial Intelligence, Self-Attention; SHAP, TON_IoT, autoencoder, deep learning, internet of things, intrusion detection system