Resource-Aware Adaptive Federated Learning for Enhanced DDoS Detection in Vehicular Ad Hoc Networks

Securing vehicular communication networks is cru-cial as increased connectivity exposes these networks to various cyber threats, particularly Distributed Denial-of-Service (DDoS). This growing vulnerability necessitates a robust framework to identify and mitigate such attacks, giving rise to distributed learning. Federated Learning has shown promise by enabling de-centralized model training across multiple clients, preserving data privacy while leveraging distributed data. However, traditional FL methods assume equal client participation in every training round, which is impractical in real-world scenarios where client resources differ. This paper introduces resource-aware federated learning (FEDRA), an adaptive federated learning strategy that dynamically selects a subset of clients based on resource states, such as battery life, CPU usage, and network connectivity. This adaptive approach optimizes the training process and enhances the global model's performance, thus improving DDoS detection. The framework was evaluated on the CICDDoS2019 dataset and the newly released CICIoV2024 dataset, demonstrating its robustness and effectiveness in inter and intra-vehicular communication scenarios.