ANOMALY-BASED INTRUSION DETECTION FOR A VEHICLE CAN BUS: A CASE FOR HYUNDAI AVANTE CN7

Abstract

Flooding, spoofing, replay, and fuzzing are common in various types of attacks faced by enterprises and various network systems. In-vehicle network systems are not immune to attacks and threats. Intrusion detection systems using different algorithms are proposed to enhance the security of the in-vehicle network. We use a dataset provided and collected in "Car Hacking: Attack and Defense Challenge" during 2020. This dataset has been realized by the organizers of the challenge for security researchers. With the aid of this dataset, the work aimed to develop attack and detection techniques of Controller Area Network (CAN) using different algorithms such as support vector machine and Feedforward Neural Network. This research work also provides a comparison of the rendering of these algorithms. Based on experimental results, this work will help future researchers to benchmark their results for the given dataset. The results obtained in this work show that the model selection does not depend only on the model's accuracy that is explained by the accuracy paradox. Therefore, for the overall result accuracy of 62.65%, they show that the support vector machine presents the most satisfying output in terms of precision and recall. The Radial basis kernel gives 65% and 67% precision for fuzzing and flooding and the recall of 64% and 100% for replay and spoofing, respectively.