A BiLSTM-Based IoT Intrusion Detection System with Mutual Information and Focal Loss

The rapid expansion of global Internet of Things (IoT) device numbers has significantly heightened the importance of securing these systems. As a core technology for IoT security protection, intrusion detection systems (IDS) have garnered significant attention from both academia and industry. This study introduces an IDS named BMF-IDS, designed to enhance IoT security. BMF-IDS leverages the advantages of bidirectional long short-term memory (BiLSTM) networks, mutual information (MI), and the focal loss (FL). Specifically, BiLSTM is utilized to identify time-based features in network traffic. MI is utilized for feature selection to reduce feature dimensions and improve the efficiency of model training and inference. The FL function addresses the problem of imbalanced network traffic datasets, enhancing detection accuracy for minority class attacks. We evaluated BMF-IDS using the Edge-IIoT dataset, and the findings indicate that BMF-IDS surpasses previous approaches in detection performance, improving detection capability while substantially reducing computational overhead.