CBMAD: Anomaly Detection in IoT Network Traffic via Consistent Bidirectional Mamba Autoencoder

The Internet of Things (IoT) has revolutionized various industries but also created substantial security vulnerabilities. Traditional security mechanisms struggle to address the evolving threats, prompting interest in machine learning (ML)-based Intrusion Detection Systems (IDS). This paper presents Consistency Bidirectional Mamba Autoencoder Anomaly Detection (CBMAD) model, a novel unsupervised multivariate time-series (MVTS) anomaly detection algorithm, for detecting anomalies in IoT network traffic. CBMAD employs a bidirectional encoder and a gate mechanism (GM) for feature extraction and fusion, coupled with a representation consistency learning framework to capture correlations among normal behaviors. Anomalous patterns are identified due to their inability to reconstruct inputs and maintain consistency across varying input perspectives. Extensive experiments conducted on several publicly available Internet of Medical Things (IoMT) network traffic datasets demonstrate that the proposed method achieves superior detection performance compared to several recent detection algorithms. These results indicate that CBMAD offers an effective solution for enhancing security in medical IoT environments.