An Ensemble Deep Learning Framework for Autonomous Vehicle Attack Detection

Recent advancements in autonomous vehicles have enhanced their functionality and connectivity by integrating programmable components and wireless communications devices such as Electronic Control Units (ECUs). However, these devices are vulnerable to cyber-attacks hence, timely detection necessary for the safety of passengers and vehicles. Several traditional rule-based systems and classical machine learning based methods for cyber attack detection have been proposed that suffer from several inherent challenges such as incapability of handling more complex or diverse threats due to the need for predefined parameters for each possible scenario and manual feature extraction process respectively. These challenges highlight the need for more advanced, and flexible methods to develop reliable security systems. To overcome challenges in the existing systems, this paper introduces a completely automated robust intrusion detection system to detect cyber-attacks in autonomous vehicles. The presented framework is trained and validated on a publicly available database and attained the highest accuracy of 94.52%, surpassing several existing architectures in performance. Additionally, the results of the proposed framework are visually interpreted using Explainable Artificial Intelligence (XAI) techniques like LIME and SHAP to showcase the framework’s transparency and trustworthiness.