Adaptive personalized federated learning with lightweight depthwise convolutional bottleneck network for novel intrusion detection system in internet of vehicles

The increasing adoption of Connected and Autonomous Vehicles (CAVs) within intelligent transportation systems has amplified concerns over cybersecurity threats in the Internet of Vehicles (IoV). To address the limitations of centralized Intrusion Detection Systems (IDS), we propose an Adaptive Personalized Federated Learning (APFed) model integrated with a Lightweight Depthwise Convolutional Bottleneck Network (LDwCBN). The system is designed to ensure privacy-preserving, resource-efficient, and accurate intrusion detection under heterogeneous and non-IID data conditions. APFed enhances model personalization and generalization through fine-grained adaptive updates and dynamic weight fusion, while LDwCBN improves detection speed and efficiency on constrained vehicular hardware. Extensive evaluations on benchmark datasets, including CIC-IDS2017, CSE-CIC-IDS2018, Car-Hacking, and CAN-Train-Test, demonstrate that the proposed method outperforms several state-of-the-art federated IDS approaches. Specifically, it achieves accuracy improvements of up to 5% over FedAvg and FedProx models, with significant gains in precision (up to 4%), recall (up to 3%), and F1-Score (up to 4%).