Towards IoT Anomaly Detection with Tsetlin Machines
Ole Gunvaldsen, Henning Blomfeldt Thorsen, Per‐Arne Andersen, Ole‐Christoffer Granmo, Morten Goodwin · 2023
Anomaly detection is pivotal in strengthening the security and reliability of Internet of Things (IoT) devices. This study delves into the capabilities of the Tsetlin Machine, an innovative machine learning algorithm, in the context of anomaly detection for IoT devices. Expanding upon existing research, we conduct comprehensive empirical investigations using five prominent network traffic datasets, including CICIDS2017, KDD99, NSL-KDD, UNSW-NB15, and UNSW Bot-IoT. Moreover, we assess the efficacy of Tsetlin Machines compared to an extensive array of algorithms that currently hold prominence in state-of-the-art network intrusion detection. Our research reveals that Tsetlin Machines emerged as a highly competitive and potent approach for anomaly detection in IoT devices, consistently delivering superior or comparable results compared to the previous methods. Notably, Tsetlin Machines afford on-device training capabilities, a distinct advantage not readily accessible with other methods. These insights suggest that the Tsetlin Machine holds significant promise as a robust and efficient tool for detecting anomalies in IoT devices, thereby strengthening IoT systems’ overall security and reliability. In summary, our paper highlights the potential of the Tsetlin Machine to address pressing IoT security challenges, paving the way for future advancements in this critical domain.