FedSplit-HIDS: A Host-Based Intrusion Detection Framework for Heterogeneous Industrial IoT Environments

This paper presents FedSplit-HIDS, a novel framework aimed at host-based intrusion detection within diverse Industrial Internet of Things (IIOT) environments. The proposed system utilizes a two-fold strategy: Apply federated learning (FL) for devices with ample resources, while incorporating split learning (SL) along with edge computing and FL for devices with limited resources. This approach effectively handles the challenge of processing varied data across multiple IIoT platforms. To enhance privacy protection, localized Rényi differential privacy (L-RDP) is applied at the device level, ensuring the security and privacy of local model parameters and gradients. FedSplit-IDS is equipped to identify a wide range of known intrusions, cyber attacks, and malware that are common in IIoT environments. We suggest that this framework outperforms existing intrusion detection systems in IIoT environments in various evaluation metrics, thus strengthening the critical security infrastructure for Industry 5.0 and integrated cyber-physical manufacturing systems.