Predicting Human Mobility with Federated Learning

In recent years, location prediction has become an important task and has gained significant attention. Existing location prediction methods rely on centralized storage of user mobility data for model training, which may lead to privacy concerns and risks due to the privacy-sensitive nature of user behaviors. In this work, we propose a privacy-preserving method for mobility prediction model training based on federated learning, which can leverage the useful information in the behaviors of massive users to train accurate mobility prediction models and meanwhile remove the need to centralized storage of them. Firstly, we propose a novel network named STSAN (Spatial-Temporal Self-Attention Network) on each user device, which can integrate spatiotemporal information with the self-attention for location prediction and a new personalized federated learning model named AMF (Adaptive Model Fusion Federated Learning), which is a mixture of local and global model. Finally, the results are superior to various baselines on four real-world check-ins datasets, verifying the effectiveness of the method.