Lightweight Privacy-Preserving IDS for ITS: Integrated Federated Learning and Blockchain

As intelligent transportation systems (ITS) become more integrated into modern infrastructure, vehicular communication systems face increasing cybersecurity threats. Traditional centralized intrusion detection system (IDS) has significant limitations in terms of the scalability, privacy preservation, and trust establishment, which are critical challenges in ITS environments. To address these issues, this paper proposes an intrusion detection method tailored specifically for ITS, integrating federated learning (FL) and blockchain technology to create a secure, scalable, and privacy-preserving threat detection system for the Internet of Vehicle (IoV). The proposed method utilizes FL for distributed model training, avoiding the sharing of raw data and employing encryption techniques to protect user privacy at the edge devices. Blockchain technology ensures the integrity and tamper-proof nature of model updates and fosters trust between entities. Moreover, to accommodate the heterogeneous and dynamic data environment in IoV, the method supports both independent and identically distributed (IID) and non-IID data scenarios, enhancing the system’s adaptability and robustness. Given the computational limitations of vehicular devices, this work incorporates knowledge distillation and lightweight model designs, effectively reducing the local computational burden. Experimental results demonstrate the significant effectiveness of the proposed approach: on the CICIDS2017 dataset, the model achieves an accuracy of 97.10% with 11,904 parameters and a memory consumption of 0.045MB; on the Car-Hacking dataset, it achieves an accuracy of 99.10% with 9,989 parameters and a memory consumption of 0.038MB; on the CICIoV2024 dataset, it achieves an accuracy of 99.65% with 9,861 parameters and a memory consumption of 0.038MB. Compared to traditional baseline models, the proposed approach significantly reduces both the number of training parameters and memory consumption, while maintaining high accuracy, making it highly suitable for deployment in resource-constrained vehicular environments within ITS.