G-NSGA3: A Dynamic Multi-Objective Feature Optimization Algorithm for Intrusion Detection in Internet of Vehicles

While providing convenience, the applications of the Internet of Vehicles (IoV) face serious threats of cyber-attacks. The detection of vehicle network attacks suffers from issues such as low detection accuracy, feature redundancy, and classification imbalance. Most existing intrusion detection studies focus on detection accuracy while neglecting the impact of the latter two factors on system performance. To overcome this limitation, an enhanced intrusion detection method utilizing Genetic Algorithm (GA) and hyperparameter optimization is proposed. This method employs GA to encode traffic features for generating an initial population, produces offspring populations through genetic operations, and selects superior individuals forming a new population using non-dominated sorting. Experiments demonstrate that this method enriches population diversity, avoids local optima, accelerates convergence speed to obtain the optimal subset, and comprehensively considers detection performance indicators such as overall accuracy, classification-specific accuracy, and feature length as fitness. This effectively enhances the performance of the intrusion detection system and optimizes the imbalance problem.