Federated learning-based intrusion detection system for edge-enabled IoT networks

The unprecedented growth of the IoT and Industrial IoT (IIoT) devices has led to severe security concerns due to which there is a need for effective intrusion detection methods to safeguard these distributed environments while also ensuring privacy-preserving techniques for data. In this paper, we have proposed a federated learning-based intrusion detection framework for IIoT using the Edge-IIoTset dataset. The suggested system includes the federated averaging (FedAvg) algorithm and the regularized deep neural network. The framework is spread among 10 clients for 30 communication rounds. The model attains 99.20%, 99.56%, 99.20%, and 99.34% accuracy, precision, recall, and F1-score, respectively, in classifying 15 attack types, including various DDoS attacks, SQL injection, XSS, ransomware, and normal traffic. The federated learning framework shows fast convergence speed. The global validation accuracy (VA) achieved 99% by round 15 (Fig. 8). At the same time, the privacy of the training data is also guaranteed, as the local data is decentralized and no raw data transmission or aggregation is required during the training process. The per-class detection rate is greater than 94% for each attack type, and the recall is 100% for MITM, Port Scanning, and DDoS HTTP attacks. The communication delay is an important indicator to verify the practicality of the proposed system. The average training time per communication round is 119.7 s, which can be run in real time in resource-constrained IIoT scenarios.