Predicting urban mobility patterns with a LightGBM-enhanced gravity model: Insights from the Wuhan metropolitan area

Abstract

Accurately quantifying urban human mobility is crucial for tackling challenges in traffic engineering, urban planning and public health. Traditional static gravity models (GModel) often fail to address spatial heterogeneity and non-linear relationships in mobility flows, particularly in complex urban regions with new towns and metropolitan areas. This study investigates mobility flows in the Wuhan Metropolitan Area by introducing an Intelligent Gravity Model (IGModel) that integrates the theoretical insights of gravity models with the non-linear predictive capacity of LightGBM. The IGModel extends the gravity model by incorporating built environment and geometric variables while leveraging machine learning to enhance flow predictions. Through this hybrid approach, the IGModel achieves robust predictive performance (R-squared = 0.97) and provides interpretable insights using SHAP (Shapley Additive Explanations) analysis. The results demonstrate the complementary strengths of mechanism-driven and data-driven approaches, with the IGModel outperforming dynamic gravity models (DGModel) and offering actionable insights for urban planning and transportation management.