Abstract

A Controller Area Network (CAN) is a vigorous bus designed for vehicle to permit\ncommunication between Electronic Control Unit (ECU) devices to transfer information and\ntogive the desired output to the driver. The CAN bus protocol uses a broadcast network without\na integral encryption mechanism, since it is designed with lack of security. Recently, it is\nrevealed that CAN bus can also be attacked remotely which can cause a major defect on the\ntraffic safety. To detect various unknown attack type that occurs during normal data\nidentification and even to detect the anomaly in general, the plan is to design a neural network\narchitecture that identifies intrusion on the CAN by monitoring CAN traffic. By using a novel\nunsupervised learning approach called GAN, both familiar and unknown intrusion scenarios\nare identified. This detection is based on deep learning system that grasp data structure of the\nhigh dimensional CAN bus in which, unique message types are passed at various time. This\nmethod can evaluate synthetic data.