Project information
- Category: Machine Learning
- Data: Campus sensor data
- Project: Work of PhD student - Xiamei
- Methods: GNN, Federated learning, Attention-based spatial-temporal GNN
- Technologies: Python, pandas, networkx, tensorflow
- Conference/Journal: IEEE ITSC Conference 2023
- URL: Proceedings needs to be published
Project details
This research presents the FedASTGNN model, which enhances data security and accuracy in traffic prediction for Intelligent Transportation Systems. By combining the federated averaging algorithm with a spatial-temporal graph neural network, this model reduces reliance on extensive datasets. Evaluation across various scenarios reveals that FedASTGNN preserves the benefits of traditional models while ensuring data confidentiality, though its performance is affected by data distribution imbalance among subnetworks. These insights highlight FedASTGNN's potential as a secure, efficient solution for traffic prediction.