An optimized lightweight DDoS detection framework towards Internet of Vehicles

The widespread deployment of the Internet of Vehicles (IoV) has significantly advanced vehicular intelligence and established a critical foundation for next-generation advanced driver assistance systems. However, this connectivity inherently expands the cyberattack surface, exposing vehicles to an escalating spectrum of sophisticated threats. In this paper, we introduce a novel framework for network attack detection that strategically leverages the inherent sparsity of attack features to achieve unprecedented detection performance. Our core innovation, the Optimized Feature Tokenizer Transformer (Opt-FT-Transformer), employs multi-head self-attention mechanisms to effectively model complex spatiotemporal dependencies within high-dimensional network traffic flows. By integrating Bayesian hyperparameter optimization, the framework maintains exceptional robustness across dynamically evolving vehicular network environments. Evaluated on the IEEE DataPort SDN-DDoS, our framework achieves 99.74% detection accuracy of 0.92% gain over unoptimized FT-Transformer, with F1-scores exceeding baselines by 4.8{7.2% under traffic dynamics. Crucially, it maintains more than 98% precision in low-sample attack regimes, vital for realtime mitigation.