DeFL-BC: Empowering Reliable Cyberattack Detection through Decentralized Federated Learning and Poisoning Attack Defense

This work presents a collaborative and trust assurance model for network attack detection using Federated Learning (FL) in edge computing environments. This study leverages the capabilities of Federated Learning (FL) and blockchain technologies, specifically the Hyperledger Fabric platform, to construct a decentralized Federated Learning model, named DeFL-BC framework that can encourage contribution from the community and resist to poisoning attacks. Our framework aims to detect and prevent network attacks effectively in the Industrial Internet of Things (IIoT) contexts with the knowledge sharing from collaborative participants without privacy leakage. The experiments conducted on two datasets, Edge-IIoTset and CIC-IDS2018 demonstrate that the proposed framework is robust in reducing a single point of failure of the system, and showing outstanding performance under poisoning attacks when the accuracy is up to more than 90% and addressing limited computing resources and intermittent network connectivity in cyberattack scenarios. Furthermore, the integration of blockchain technology enhances security and resilience, effectively mitigating the risks of the single point of failure in conventional FL approaches.