[4Xin2] Poster session 2

[4Xin2-59] Unsupervised Anomaly Detection using VQ-VAE and Transformer considering Statistics and Sequence of Flow Data

〇Norihiro Okui¹, Shotaro Fukushima², Ayumu Kubota¹, Takuya Yoshida³ (1.KDDI Research, Inc., 2.ARISE analytics, inc., 3.Toyota Motor Corporation)

Keywords:Anomaly detection, Cyber Security, IoT

The widespread use of IoT devices has increased the threat of cyber-attacks, making anomaly detection even more critical. Flow data can reduce the amount of data for analysis and is one of the promising data formats in anomaly detection. However, improving detection accuracy is challenging since flow data contain less information than packets. Recently, anomaly detection methods based on natural language processing techniques have been proposed, and improved accuracy has been reported by considering the sequential features of the flow data. Nonetheless, the statistical information critical for anomaly detection is lost when discretizing each flow. In this study, we focus on the sequence of flow data and propose a novel anomaly detection method utilizing NLP techniques combined with VQ-VAE, which automatically quantifies traffic data. Experimental results on the ToN-IoT dataset show that the proposed method's ROC-AUC is 0.688 and higher than that of previous studies.