Intrusion Detection For Controller Area Network Using Support Vector Machines

Controller Area Network is the most widely adopted communication standard in automobiles. The CAN protocol is robust and is designed to minimize overhead. The lightweight nature of this protocol implies that it can't efficiently process secure communication. With the exponential increase in automobile communications, there is an urgent need for efficient and effective security countermeasures. We propose a support vector machine based intrusion detection system that is able to detect anomalous behavior with high accuracy. We outline a process for parameter selection and feature vector selection. We identify strengths and weaknesses of our system and propose to extend our work for time-series based data.