Dynamic Federated Learning Aggregation for Enhanced Intrusion Detection in IoT Attacks

The widespread integration of Internet of Things (IoT) devices has elevated security risks, specifically through vulnerabilities exploited by Botnet attacks. The emergence of these attacks highlights the necessity for resilient intrusion detection systems to detect such security threats. Prior methods employed a conventional centralized approach, involving the collection of data followed by training a machine learning or deep learning model. However, the conventional methods prove impractical for enterprise entities, as they are reluctant to share their sensitive data in a centralized environment due to privacy concerns. In this study, we utilize a decentralized Federated Learning (FL) approach to detect such Botnet attacks (i.e., Bruteforce, DDoS, DoS, Infiltration and Bot attacks). Our proposed FL method incorporates dynamic aggregation, facilitates the aggregation of updated models at local clients, thereby addressing the limitations associated with centralized data handling and privacy concerns. Our results show that the proposed method has achieved 87.98% accuracy for Botnet attacks and 100% accuracy for DoS and DDoS attacks.