Comparative Study of Ensemble Learning Techniques for Fuzzy Attack Detection in In-Vehicle Networks

Nowadays, vehicles have become more complex due to the increased number of electronic control units communicating through invehicle networks. Controller area network (CAN) is one of the most used protocols for in-vehicle networks. Still, it lacks a conventional security infrastructure, making it highly vulnerable to numerous attacks. The Fuzzy attack is one of the most challenging attacks for in-vehicle networks because of its randomly spoofed injected messages similar to the legitimate traffic and its numerous physical effects on the vehicle. In this paper, we focus on Fuzzy attack detection in the internal vehicle network by investigating the performances of ensemble learning techniques to mitigate this attack. We evaluated their efficiency on realistic datasets and on a new advanced stealthy attack dataset with physical impacts on the vehicle. eXtreme, Light, and Category Gradient Boosting, as well as Bagging ensemble learning techniques, in particular, showed a considerable improvement in detection performance in terms of accuracy, training and testing time reduction, and a decreased false alarm rate.

Automated summary

This paper investigates the detection of Fuzzy attacks in the internal vehicle network using ensemble learning techniques. The efficiency of these techniques was evaluated on realistic datasets and a new advanced stealthy attack dataset with physical impacts on the vehicle. The results show that eXtreme, Light, and Category Gradient Boosting, as well as Bagging ensemble learning techniques, significantly improve the detection performance in terms of accuracy, training and testing time reduction, and decreased false alarm rate.