A Privacy-Preserving Federated Intrusion Detection Framework with Reputation-Aware Client Selection and Integrity Verification

Abstract

Abstract With the increasing number of Industrial Internet of Things (IIoT) networks, critical infrastructures are now more vulnerable to cyberattacks. In this context, the need for distributed and privacy-preserving intrusion detection systems has become essential. In this paper, we introduce a secure federated learning framework for intrusion detection in IIoT networks that supports model training in non-IID environments without sharing raw data. In this system, each client maintains a lightweight MLP model locally, and a client-level DP-SGD is used to enhance privacy and hashing to maintain update integrity. Also, to consciously select clients and reduce the impact of malicious clients, a reputation-based mechanism is proposed that leverages the ideas of trust management in blockchain, but can be implemented without the need for a full blockchain implementation. The performance of the proposed model on the Edge-IIoTset dataset in binary and multi-class classification and in 3, 5, and 7 clients shows that the proposed model achieves an accuracy of over 98% in all scenarios, which is close to the results of the centralized approach.