Intrusion Detection in the Internet of Vehicles Using Transformer Models

The proliferation of Internet of Vehicles (IoV) systems has introduced critical cybersecurity vulnerabilities in connected vehicle infrastructures. This paper presents a novel application of Transformer architecture for intrusion detection in vehicular Controller Area Network (CAN) buses, specifically addressing Denial-of-Service and Spoofing attacks. We introduce a Transformer-based model optimized for CAN frame sequence analysis, evaluated on the CICIoV2024 dataset comprising real-world attack scenarios from a 2019 Ford vehicle. Our experimental results demonstrate superior detection capabilities compared to classical machine learning approaches and recurrent neural architectures, achieving 98.4% accuracy and 98.5% F1-score. The proposed architecture's self-attention mechanism effectively captures temporal dependencies in CAN frame sequences while maintaining computational efficiency (2.3ms inference time) suitable for automotive-grade ECUs. This research advances the state-of-the-art in IoV security through enhanced detection accuracy and practical deployment considerations for resource-constrained vehicular environments.