CAN Bus Intrusion Detection System Using Light Gradient Boosting Machine

The in-vehicle controller area network (CAN) is the nervous system of autonomous vehicles that allows the electronic control units (ECUs) to communicate without much complexity. Before the invention of CAN, ECUs were connected through complex wiring to communicate. With the introduction of CAN, ECUs can now send messages by being connected through the CAN serial bus. However, CAN is susceptible to threats. An adversary can get access to the CAN bus by exploiting vulnerabilities in the car’s internal interfaces such as Bluetooth and WiFi. The adversary may transmit malicious instructions via the bus, such as preventing a vehicle from breaking when necessary, resulting in accidents. In this project, we propose an anomaly detection system based on stacked Convolutional Neural Network and Light Gradient Boosting Machine (CNN - LightGBM) algorithms to detect CAN bus attacks. We will train and test our detection approach using an actual vehicle’s CAN attack datasets, which contain DoS attacks, fuzzy attacks, and spoofing attacks. Furthermore, we will evaluate the efficiency of our approach through different standard metrics such as recall, precision, F1 score, accuracy, and false-positive rates (FPR). The analysis of the proposed model achieved an overall accuracy of over 99 percent.