Cost-Sensitive Detection of DoS Attacks in Automotive Cybersecurity Using Artificial Neural Networks and CatBoost

Connected automotive components are increasingly vulnerable to cyberattacks, particularly if they lack adequate protection. The controller area network (CAN), the most widely used communication protocol in vehicle systems, inherently lacks crucial security mechanisms to safeguard data transmission. Enhancing cybersecurity in the automotive sector is therefore vital to detect and prevent potential security breaches. Our research focuses on one of the most common attack types in automotive environments: Denial-of-Service (DoS) attacks, which aim to disrupt service availability. To address this challenge, we present a novel approach that integrates cost-sensitive learning and Optuna-driven hyperparameter optimization to significantly improve the detection of DoS attacks on the CAN network. By incorporating cost-sensitive learning, we tailor the model to minimize the cost of misclassifications, particularly False Negatives, which are critical in cybersecurity contexts. Additionally, Optuna is used to optimize the model's hyperparameters, ensuring robust and efficient performance. Our proposed framework leverages a pipeline that combines artificial neural networks (ANN) with CatBoost, demonstrating superior performance in detecting DoS attacks. The model achieves outstanding results, as evidenced by high precision, recall, F1-scores, and an impressive AUC-ROC score of 99.99%.