Intrusion Detection System on IoT Devices via Federated Learning

In the rapidly changing Internet of Things (IoT) landscape, IoT devices that control our everyday lives are becoming a primary target for intruders. It has tended to become more vulnerable and exposed to handling a diverse range of data and functions. Traditional centralized intrusion detection systems failed to address its robustness against the scale and diversity of IoT networks. Furthermore, centralized ML models pose significant privacy concerns because sensitive data transfers to a central server, which opens up potential data leakage. To address these challenges, we propose Federated Learning-based Intrusion Detection System (FLIDS). Our FLIDS not only preserves privacy but also minimizes the resources required for data transmission. With FLIDS, real-time learning and adaptation to new attacks are always possible because IoT devices can continuously update their models based on local changes in network data without waiting for batch processing, which is typically required in traditional ML models. Moreover, FLIDS has been implemented using Federated Proximal to address the heterogeneous data environment, and different training configurations have been performed to tune the model parameters. Our model achieved a higher accuracy of 99.31% and an f1-score of 97.38%, underscoring the excellent detection performance.