Addressing the Privacy and Complexity of Urban Traffic Flow Prediction with Federated Learning and Spatiotemporal Graph Convolutional Networks

Traffic flow prediction is significant for metropolitan life today. However, existing systems predominantly adopt a deep learning model, often falling short of adequately safeguarding user privacy. Moreover, these systems tend to overlook how external factors affect traffic flow. To tackle these concerns, we propose a novel architecture based on federated learning and Spatiotemporal GCN. Simultaneously, we employ graph embedding techniques to incorporate external factors into the road network, which helps the model to consider multiple factors affecting traffic flow more fully. Evaluation on the real dataset shows that our framework can achieve high accuracy while preserving privacy.