A Lightweight Intrusion Detection for Attacks in IoT Big Data Networks

Abstract

Industrial Internet of Things (IoT) security vulnerabilities create a critical challenge that needs to be addressed. Detecting intrusion from huge amounts of data in IoT Big Data networks is challenging. Deep learning models have been used to address this challenge. However, false negatives still create increased vulnerabilities. Motivated by this gap, this paper introduces a machine-learning intrusion detection model with a two-layer feature selection stage. Combining SHaply Additive exPlanations and autoencoder as a two-layer feature selection method provides a comprehensive feature selection and dimensionality reduction.  The model evaluation results demonstrate that utilizing a two-layer feature selection improves the performance of the intrusion detection model compared to a one-layer-based feature selection found in previous works while enhancing processing speed and feature interpretability.