GAN-augmented adversarial training for robust detection of complex jamming attacks in VANET intelligent systems

Abstract

Vehicular Ad-hoc Networks (VANETs) are crucial for enabling real-time communication in intelligent transportation systems. While network availability and safety-critical services cannot be compromised under normal circumstances, they are highly vulnerable to jamming attacks. Jamming detection through traditional machine learning models is mainly static, and models are trained offline, making them prone to adaptation when facing various jamming patterns or adversarial updates. In this work, we introduce a robust and adaptive mechanism featuring Generative Adversarial Networks (GANs) for realistic attack simulation, a hybrid Convolutional Neural Network–Random Forest (CNN-RF) classifier for robust detection, and an online learning process with concept drift adaptation. Under the augmentation of the GAN module, adversarial and zero-day jamming behaviors are generated as new samples to expand the dataset; at the same time, the classifier is trained in an adversarial manner to ensure performance without damage under perturbation. Experimental results on a real-world DSRC vehicular intelligent dataset show that the proposed model achieves a 93.4% F1 score while reducing inference latency by 44.7% and attaining 92.3% model accuracy with over 10% drift—outperforming all other state-of-the-art baselines by a significant margin. These results demonstrate the potential of the model for real-time resilience and adversarial deployment within dynamic VANET settings.