Towards Interpretable and Lightweight Intrusion Detection for In-vehicle Network

Along with the improvement of informatization and intelligence in modern vehicles, the attacks aimed at in-vehicle network has been becoming an emerging threat. Current methods such as neural network-based methods have shown a state-of-the-art performance on intrusion detection based on in-vehicle network monitoring. However, the limitations in interpretability and resource consumption remain the non-negligible obstales for practical applications and deployments. In this paper, we propose and build an interpretable and lightweight framework based on shapelets model for in-vehicle network intrusion detection. We conduct massive experiments on a real-world dataset, namely Car-Hacking 2019 to evaluate our detection framework. Experiments demonstrate that our proposed method can reach or surpass state-of-the-art methods in four different adversarial scaneraios with extremely low time consumption, while providing interpretable results with visual explanations for further investigations.