Performance Comparison of Machine Learning and Federated Learning for Intrusion Detection in Vehicular Ad-Hoc Networks

The proliferation of connected autonomous vehicles (CAVs) underscores the necessity for cybersecurity protocols. Cyberattacks on CAVs may result in data breaches and the manipulation of vehicle control, potentially leading to accidents or safety risks. Federated learning (FL) decentralizes machine learning (ML) by training remote-device models while maintaining data locality. Although it promises privacy-sensitive domains such as cybersecurity, it encounters obstacles, including data heterogeneity, communication overhead, and model convergence issues. This study investigates whether models that perform well in centralized ML maintain their performance in FL. The results reveal that strong centralized ML performance does not always translate to FL. The findings highlight the importance of optimizing FL models to handle limited local data effectively.