Comparative Study of Machine Learning Methods for In-Vehicle Intrusion Detection

An increasing amount of cyber-physical systems within modern cars, such as sensors, actuators, and their electronic control units are connected by in-vehicle networks and these in turn are connected to the evolving Internet of vehicles in order to provide ""smart"" features such as automatic driving assistance. The controller area network bus is commonly used to exchange data between different components of the vehicle, including safety critical systems as well as infotainment. As every connected controller broadcasts its data on this bus it is very susceptible to intrusion attacks which are enabled by the high interconnectivity and can be executed remotely using the Internet connection. This paper aims to evaluate relatively simple machine learning methods as well as deep learning methods and develop adaptations to the automotive domain in order to determine the validity of the observed data stream and identify potential security threats.