Securing smart city Internet of Vehicles via transformer integrated consensus ensemble learning and bio-inspired metaheuristics

The Internet of Vehicles (IoV), an extension of the Internet of Things (IoT) concept, facilitates online communication and connectivity among smart vehicles. The advanced features of smart vehicles have drawn customer interest in electric vehicle technology. The rapid development of the Internet of Vehicles (IoV) raises important privacy and security issues that could lead to dangerous events. Many researchers have developed deep learning based systems for IoT network intrusion detection. These models aim to reduce smart vehicle accidents and identify compromising network attacks. In this paper, a consensus-driven ensemble of classifiers, Ant Colony Optimization (ACO) for feature selection, and vision transformer-based feature extraction are all included in the proposed system. Initially, the CLIP vision transformer model is used to extract semantic features from vehicle network data. After that, ACO selects the best feature subset to increase accuracy and decrease complexity. Predictions are integrated using a consensus ensemble of Support Vector Machine (SVM), K Nearest Neighbor (KNN), and Logistic Regression (LR), which selectively applies stacking to improve multiclass intrusion detection. The evaluations were conducted using two data sets: the CICEVSE dataset, which contains 22,086 samples from eight different intrusion categories, and the publicly available Car Hacking dataset, which contains 29,228 samples from five different intrusion categories. The experimental results demonstrate that the proposed approach achieved a maximum score of 100% on the Car Hacking dataset and 99.29% on the CICEVSE dataset, reflecting optimum accuracy.