Multi-Vehicle Behavioural Framework for Proactive Threat Hunting in Heterogeneous In-Vehicle CAN Networks for Consumer Electronics

Existing Controller Area Network (CAN) intrusion detection systems (IDSs) are mostly passive, vehicle-centric, and susceptible to dataset bias, resulting in limited robustness in Multi-vehicle deployments and realistic consumer electronics environments. This paper introduces a Multi-vehicle behavioural model to proactively detect threats in vehicular CAN networks, incorporating transformer-based temporal modelling, diffusion-based latent refinement, defence implemented via reinforcement learning, and explainable AI. The framework trains vehicleagnostic behavioural representations from CAN traffic, balances Multi-domain generalisation through latent regularisation, and enables proactive cyber defence via a Deep Q-Network (DQN). Integrated Gradients are also used to provide interpretable detection information for safety-critical settings. The proposed methodology is evaluated on the multi-vehicle UIDs-CAN, HCRL, and CICIOV2024 benchmarks under time-safe and Multi-vehicle conditions. Experimental findings indicate strong detection, with an F1-score of 99.86% and an ROC-AUC of 100%. Moreover, CICIOV2024 achieves 100% accuracy and real-time inference with a mean latency of 1.58ms per sample. These findings suggest that explainable, behavioural, and proactive IDS frameworks can significantly enhance the security of CAN intrusion detection on heterogeneous vehicles for the consumer electronics ecosystem.