Federated Learning Models for Intrusion Detection in Industrial IoT Networks

The Industrial Internet of Things has emerged as an essential tool for building Industry 4.0 and Industry 5.0 where timely information can be retrieved from different scenarios. These devices are highly vulnerable to cyberattacks as heterogeneous types of devices can be used in the infrastructure that may or may not be equipped with standardized security protocols. Artificial intelligence-based methodologies can effectively identify these types of attacks on a prior basis for taking mitigation action. This method raises concerns about data privacy as building a machine learning-based method requires the sharing of network data that can reveal the actual information of industries. The proposed Federated Learning based framework handles this concern by preserving each device’s critical data by utilizing the benefits of the decentralized model aggregation. This research work presents the comparison of the proposed framework on the CICIoT2023 dataset with federated averaging and federated proximal techniques for achieving a global model. The performance evaluation of these two aggregation techniques is performed based on metrics of accuracy, loss, precision and recall. The results prove that the federated proximal method achieves higher accuracy in comparison to the federated averaging method.