Enchanced CAV Security Using Machine Learning

As Connected and Autonomous Vehicles (CAVs) have become increasingly popular over the past few years, more people rely on these vehicles to offer better safety features and efficient transportation options. However, their reliance on interconnected systems makes them vulnerable to various cyber-attacks. This poses an issue to car manufacturers and individuals who purchase these vehicles as they are susceptible to many digital attacks. Furthermore, CAVs are becoming more common, so understanding and addressing their cyber-attack vulnerabilities is vital. Our plan is to improve CAV security by creating new and improved defense mechanisms against cyber threats. The HCRL-Car Hacking Dataset provided us with many Controller Area Network (CAN) messages. These messages comprised normal and injected messages simulating attacks. It also provides us with data from some of the common types of cyber-attacks such as DoS and Fuzzy. We implemented 4 ML models to provide the attack classification, including K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and Decision Trees. Each model was implemented to detect anomalies indicative of the types of attacks focused in the dataset. Various data preprocessing steps were executed as the dataset contained hexadecimal values and missing values. Furthermore, the dataset was split into training and testing datasets with a ratio of 70% to 30% respectively. Additionally, we employed feature selection to identify which features would present the highest correlation and feature importance score. Using the results from feature selection, we applied StandardScalar to assist those features in identifying cyber-attack patterns on CAVs. We achieved the best results from the KNN and SVM of 99.5% and 98% respectively. These results improve the security of cyber-physical systems by suggesting a useful strategy to protect CAVs from cyber-attacks.