Edge Computing-Based Distributed Intrusion Detection Systems via Multi-Hop Split Learning

Intrusion Detection Systems (IDS) play a critical role in monitoring network traffic and protecting against cyber threats. As networks evolve toward beyond 5G (B5G) or 6G, edge computing presents new opportunities for real-time distributed intrusion detection. We investigate and propose a distributed IDS framework based on multi-path multi-hop learning, specifically designed for an edge computing environment by partitioning a deep learning model to spread over multiple physical nodes to operate. This architecture supports collaborative model training among multiple edge nodes, while ensuring user data privacy and overcoming the limitations of centralized systems. To validate system performance, we performed extensive experiments on three benchmark datasets (ToN_IoT, IoT-23, and UNSW-NB15). Experimental results show that compared to a centralized IDS, our edge computing-assisted distributed IDS achieves significant reductions in training time, up to 44.4% faster, while maintaining competitive detection performance. These improvements are particularly important for real-time applications of B5G/6G networks. Our simulation results validate the effectiveness, scalability, and practicality of partitioned learning-based IDS when deployed at the network edge.