CANPerFL: Improve In-Vehicle Intrusion Detection Performance by Sharing Knowledge

The Controller Area Network (CAN) is a widely used communication protocol in automobiles, but it vulnerable to various types of attacks. To address this issue, researchers have been exploring the use intrusion detection systems (IDS) for CAN bus. Deep learning and machine proven be powerful tools detecting intrusions accurately quickly. However, deep models require large amounts data achieve optimal performance, which can challenging case bus IDS. overcome challenge, we propose novel learning-based IDS called CANPerFL that uses personalized federated scheme aggregate datasets from different car models. By building universal model trained on small amount each manufacturer, provide global knowledge transferred improve performance participant. demonstrate efficiency proposed model, collected real dataset consisting three models: KIA, BMW, Tesla. experimental results show increases F1 scores by 4% overall, compared baselines. Moreover, system provides significant advantages when local participant relatively small. According our experiments, more than 90% with at least 30k training samples client. Finally, empirically takes benefits joining system.