Comparative Analysis of Federated Learning Techniques in Big Data-Driven Intelligent Transportation Systems

The advancement of Intelligent Transportation Systems (ITS) represents a pivotal development in IoT technology, improving traffic management, reducing accidents, and optimizing transport networks. ITS generates vast data streams from interconnected devices, offering valuable insights into travel and road conditions. However, leveraging this extensive data poses significant performance, privacy, and scalability challenges. In this context, Federated Learning (FL) emerges as a crucial solution. This paper contributes a comprehensive comparative analysis of various FL algorithms in processing ITS Big Data, focusing on their effectiveness in managing and analyzing data while preserving user privacy and ensuring scalability. Using a dataset from Udacity, provided by Roboflow, consisting of 15,000 high-quality images across 11 categories formatted for YOLO, we conduct a comparative evaluation of FedSGD, FedAvg, and FedProx. Through this comparative evaluation, we demonstrate the potential of FL to manage and analyze ITS data while preserving user privacy and ensuring scalability. Our results highlight that FedAvg achieves the highest accuracy at 79.01%, outperforming FedSGD at 78.10% and FedProx at 78.61%.