Hybrid Feature Selection for Efficient Machine Learning-Based Intrusion Detection in IoT Networks
Mohammed Nagah Amr, Tamer Mekkawy, Ashraf M. Mahran, Ahmed S. Elliethy · 2025
The expansion of Internet of Things (IoT) infrastructure has increased data dimensionality, creating challenges for intrusion detection systems (IDS) in resource-constrained environments. Lightweight IDS solutions, focusing on reduced feature sets, have become essential to minimize computational complexity and ensure efficient deployment on IoT devices. This work proposes a novel hybrid feature selection technique for IDS that aggregates two filter-based methods to identify highly relevant, non-redundant, and interaction-friendly features. Specifically, the approach combines Joint Mutual Information (JMI) for selecting non-redundant features and the XGBoost weight feature importance score for capturing complex feature interactions. Then, a mathematical set intersection is applied, resulting in a reduced yet highly effective feature subset. Extensive experiments were conducted on three diverse IoT datasets, covering both binary and multi-class classification scenarios. The results demonstrate that the proposed method outperforms state-of-the-art feature selection techniques, achieving superior accuracy across most classifiers and datasets, while maintaining a lower False Positive Rate (FPR).