A Lightweight Intrusion Detection Model for In-vehicular CAN Networks

The Intelligent Transport System(ITS) is an important development in this technological era. Effective communication in ITS is possible by means of different network systems and technologies, and as a result, driverless vehicles are becoming the target of cyber-attacks. The CAN bus is mainly used in ITS for internal communication and is vulnerable to cyber-attacks. CAN protocol gives minimal information for the design of intelligent algorithms for detecting intruders. The existing intrusion detection mechanisms are computationally expensive and may not be suitable for low-end ECUs in vehicles. In this work, we have given importance to the data part of the frame while preparing the dataset for the experiments. We have explored the performance of several classical machine learning algorithms like Random Forest algorithm, XGBoost algorithm, LightGBM algorithm, Naive Baye’s algorithm, and Decision Trees with the refined dataset and used the publically available dataset ’Car Hacking: Attack and Defence Challenge’ for the experimental evaluation. In the results, the Random Forest algorithm got the most significant accuracy with 95% and an F1-score of 95%. Compared to the computationally complex algorithms, we could achieve comparable and significant results with the classical machine learning algorithms which can be easily portable to ECUs in vehicles.