UST-GNN: A Unified Spatial--Topological Graph Neural Network Framework for Urban Analytics--Demonstrated through a Case Study on Urban Health Prediction
Researchers have developed UST-GNN, a novel framework that unifies spatial and topological graph neural networks for urban analytics. This approach integrates neighborhood connectivity, diverse urban features, and location embeddings to better understand complex urban outcomes. Demonstrated through a case study on urban health prediction in Greater London using the MedSAT dataset, UST-GNN significantly outperformed existing methods, improving out-of-sample R^2 by up to 13.2%. The framework also includes a module for interpreting learned embeddings, offering insights into policy-relevant factors and aiding in urban planning and decision-making. AI
IMPACT This unified framework could improve predictive accuracy and interpretability in urban planning and public health initiatives.