A Dynamic Approach for Detecting Attacks in Controller Area Networks

Over the past few years, the interest in autonomous vehicles has significantly grown due to the ease they provide in our day-to-day lives. CAN bus is one of the most widely utilized standards for communication between Electronic Control Units (ECUs). As a result of its widespread usage, it is vulnerable to different cyber security attacks. In this work, we propose a new ensemble model that uses a heuristic rule-based approach combined with the advantages of Deep Learning techniques. This method uses Recurrent Neural Network (RNN) frameworks to create a learning system that improves detection accuracy and resilience against various attacks. The framework shows promising results with precision, accuracy, recall, and f1-score of 98%. Furthermore, the proposed model took an average of 0.0061s to detect each attack, outperforming prior intrusion detection approaches. Furthermore, we have provided multiple visual representations to better understand and validate the anomaly detection by the proposed system. Our algorithm is designed to be computationally efficient making it suitable for real-time environments where resource constraints are considered. The proposed framework provides a scalable and adaptable model for future advancements in vehicular network security.