An Intelligent Data-Driven Model to Secure Intra-Vehicle Communication Based on Machine Learning

Abstract

Modern vehicles contain 70–100 Electronic Control Units (ECUs) interconnected via the Controller Area Network (CAN) bus—a protocol engineered for real-time reliability but not for cybersecurity. The absence of authentication and encryption on the CAN bus renders intra-vehicle communication susceptible to message injection, spoofing, and denial-of-service attacks, with potentially life-threatening consequences. This paper presents an intelligent, data-driven intrusion detection system (IDS) that leverages machine learning to monitor and secure intra-vehicle CAN bus communication. Three models—Random Forest, a one-dimensional Convolutional Neural Network (CNN), and a Long Short-Term Memory (LSTM) network—were trained and rigorously evaluated on the publicly available HCRL Car Hacking dataset, which contains four attack categories: DoS, Fuzzy, RPM Spoofing, and Gear Spoofing. The LSTM achieved the highest overall accuracy of 99.4% and F1-score of 0.97, while Random Forest offered sub-millisecond inference with 97.8% accuracy, making it viable for resourceconstrained ECU deployment. A comprehensive per-attack analysis reveals that DoS attacks are detectable by all three models, whereas stealthy spoofing attacks demand sequential deep learning approaches. The proposed non-intrusive pipeline introduces no modification to existing CAN hardware, operates within strict realtime constraints, and establishes a foundation for production-grade automotive cybersecurity systems.