Lightweight Privacy-Preserving Federated Deep Intrusion Detection for Industrial Cyber-Physical System

다국어 초록 (Multilingual Abstract)

The emergence of Industry 4.0 entails extensivereliance on Industrial Cyber-Physical Systems(ICPS). ICPS promises to revolutionize industries byfusing physical systems with computational functionality.
However, this potential increase in the use of ICPSmakes them prone to cyber threats, necessitating effectivesystems known as Intrusion Detection Systems (IDS). Theprovision of privacy, system complexity, and system scalabilityare major challenges in IDS research. We presentFedSecureIDS, a novel lightweight Federated Deep IntrusionDetection System that combines CNNs, LSTMs,MLPs, and Federated Learning (FL) to overcome thesechallenges. FedSecureIDS solves major security issues,namely eavesdropping and Man-in-the-Middle attacks,by employing a simple protocol for symmetric session keyexchange and mutual authentication. Our Experimentalresults demonstrate that the proposed method is effectivewith an accuracy of 98.68%, precision of 98.78%, recallof 98.64%, and an F-score of 99.05% with different edgedevices. The model is similarly performant in conventionalcentralized IDS models. We also carry out formalsecurity evaluations to confirm the resistance of theproposed framework to known attacks and provisioningof high data privacy and security.