ML-based Anomaly Detection for Intra-Vehicular CAN-bus Networks

With advancements in the automotive industry and self-driving cars, automotive cybersecurity has become a pivotal and prioritized issue for all automakers. The CAN-bus is widely used to exchange data between vehicular networks and components, such as safety-critical systems and infotainment. Despite its importance, CAN bus systems lack security safeguards, exposing them to various security threats. This work proposes a novel two-stage Anomaly Detection System that detects malicious attacks with high detection accuracy incurring low latency. The novelty of this work is to use machine learning techniques like Decision Tree, Random Forest, and XGBoost to get predictable results while balancing with rule-based systems and keeping computational effort low. The experimental results demonstrate that our proposed model can detect different attacks with high accuracy (above 90%) on the CAN-intrusion-dataset with very low execution time and high F1-score. Thus, proving the effectiveness and efficiency of our hybrid model.