FedAGAT: Real-time traffic flow prediction based on federated community and adaptive graph attention network

Rasha Al-Huthaifi, Tianrui Li, Zaid Al-Huda, Chongshou Li

Abstract: Highlights•Introducing FedAGAT: A model balancing accuracy, speed, and privacy in traffic flow prediction.•Improved spectral community detection boosts federated learning in global-local road networks.•Enhanced federated learning scalability with FedAVG and random sub-sampling for traffic flow.•FedAGAT outperforms 7 models in traffic prediction using METR-LA, PEMS-BAY with lower costs.•FedAGAT shows superior training and inference times against DCRNN, GMAN, ASTGAT baselines.