TOFL: Time Optimized Federated Learning

Vehicular networks face cyber threats that can harm drivers, passengers, and pedestrians. In this scenario, federated learning is a possible solution to train models that detect threats without violating user privacy. However, federated learning is particularly sensitive to communication delays, which is a natural consequence of high mobility in vehicular networks. This problem is commonly ignored in the literature, which does not consider the possibility of network disconnections. This work proposes a client selection strategy designed to minimize the training time of a machine learning model for vehicular threat detection, considering the communication time that varies according to the movement of clients. The results demonstrate that TOFL, using only 20% of the total available clients, can reduce the time required to achieve high accuracy by up to 50% compared to state-of-the-art approaches, while reducing the resource consumption of client devices.