eliesgherbi/Deep-Learning-4-IDS
Modern and future vehicles are complex cyber-physical sys-tems. The connection to their outside environment raises many securityproblems that impact our safety directly. In this work, we propose a DeepCAN intrusion detection system framework. We propose a multivariatetime series representation for asynchronous CAN data. This represen-tation enhances the temporal modelling of deep learning architecturesfor anomaly detection. We study different deep learning tasks (super-vised/unsupervised) and compare different architectures, to propose anin-vehicle intrusion detection system that fits constraints of memory andcomputational power of the in-vehicle system. The proposed intrusiondetection system is time window wise: any given time frame is labelledeither anomalous or normal. We conduct experiments with many types ofattacks on an in-vehicle CAN using SynCAn dataset. We show that oursystem yields good results and allow to detect different kinds of attacks.