Applying deep learning techniques to can bus attacks for supporting identification and analysis tasks

Cars are no longer only mechanical vehicles. As a matter of fact, they contain an ecosystem of several electronic units able to exchange data using the serial communication provided by the CAN bus. CAN packets are broadcasted to all components and it is in charge of the single component to decide whether it is the receiver of the packets, in addition the protocol does not provide source identification of authentication: these are the reasons why the CAN bus is exposed to attacks. In this paper we design a method to identify CAN bus targeting attacks. The proposed method takes into account deep learning algorithms i.e., the Neural Network and the MultiLayer Perception. We evaluated our method using CAN messages gathered from a real vehicle injecting four different attacks (i.e. dos, fuzzy, gear and rpm), obtaining encouraging results in attacks identification.