Interpretable Anomaly Detection in Industrial Control Systems Using Federated Learning
Laxmi Thodupunuri, Nishanka Peesari, Sairam Utukuru · 2024
For IoT devices to be reliable and secure, anomaly detection is essential, but conventional methods often struggle to handle the distributed data of IoT. Another key challenge is the lack of privacy protection in centralized anomaly detection approaches, which require data to be gathered and analyzed in a central location. Federated Learning (FL) provides a way to train models collaboratively across decentralized Internet of Things devices while protecting the privacy of user data. In this work, an FL-based anomaly detection framework for IoT devices has been proposed, leveraging the FedAvgM algorithm to train an Autoencoder model. This works experimental results demonstrate the framework’s effectiveness in detecting anomalies in IoT traffic data and maintaining data privacy. This work demonstrates how FL can improve IoT system security and dependability while resolving privacy issues related to conventional anomaly detection techniques. By leveraging FL, proposed method has achieved better accuracy than traditional methods while also preserving sensitive IoT data.