Nation-wide human mobility prediction based on graph neural networks.

Area of Science:

• Data Science
• Urban Planning
• Network Science

Background:

• Human mobility flow prediction is crucial for urban planning and epidemiology.
• Existing methods primarily focus on intra-urban (within-city) travel forecasting.
• There is a need for models that can predict inter-urban (between-city) mobility at a national scale.

Purpose of the Study:

• To develop a nation-wide mobility predictor for anticipating inter-urban displacements.
• To utilize Graph Neural Networks (GNNs) for capturing latent relationships between geographical regions.
• To create a single, unified model for processing mobility data across diverse areas.

Main Methods:

• A Graph Neural Network (GNN) architecture was employed to model spatial dependencies.
• The model was trained and evaluated using an open dataset of nationwide trips in Spain.
• Weather conditions were incorporated as a feature in the prediction model.

Main Results:

• The proposed GNN predictor demonstrated high accuracy in forecasting the number of inter-urban trips.
• The model achieved reliable predictions across multiple time horizons.
• A significant advantage was the model's ability to process all mobility areas with a single unified network.

Conclusions:

• The developed GNN approach effectively predicts nation-wide human mobility flows.
• This method offers a more efficient and scalable solution compared to area-specific models.
• The findings support the application of GNNs for large-scale mobility forecasting in diverse domains.