A lightweight hybrid approach for intrusion detection systems using a chi-square feature selection approach in IoT

Abstract

Protecting IoT networks from cyber threats is challenging, especially with resource-constrained devices. This paper proposes an efficient, lightweight hybrid intrusion detection system (IDS) specifically optimized for IoT devices. Our innovative approach integrates convolutional neural networks (CNN) for effective spatial feature extraction and bidirectional long short-term memory (BiLSTM) networks for capturing temporal dependencies. Crucially, we employ a chi-square (${\chi }^2$) feature selection method, significantly reducing input complexity by selecting the 20 most relevant features from the UNSW-NB15 dataset. Benchmarking against recent IDS methods, our model achieved outstanding accuracy: 97.90% for binary classification and 97.09% for multiclass scenarios, clearly outperforming existing approaches. Additionally, computational performance evaluation reveals rapid prediction times (1.1 s binary; 2.10 s multiclass), demonstrating suitability for real-time IoT deployment. This study illustrates a balanced trade-off between high accuracy and low computational demand, highlighting the practical benefits of advanced, resource-efficient IDS solutions for IoT security.