Improve GWR with Explainable Machine Learning: A Case Study on Factors Influencing Human Mobility

Keywords: GWR; Explainable Machine Learning; SHAP; Human Mobility
Abstract Type: Paper Abstract

Abstract

Geographically Weighted Regression (GWR) has proven to be a powerful tool for understanding spatial heterogeneity in relationships between dependent and independent variables. This study presents an innovative methodological framework that integrates Explainable Machine Learning, SHapley Additive exPlanations (SHAP), and GWR to improve spatial data analysis. Traditional GWR models, which rely on simple linear regression, often struggle to capture the complexity and nonlinear nature of spatial data. By leveraging Machine Learning for model selection and hyperparameter tuning, we enhance the predictive accuracy and flexibility of the GWR model, while SHAP provides a layer of interpretability by explaining variable contributions at both global and local scales. The framework is applied to examine the spatial relationships between socio-economic, environmental, and land-use factors and human mobility in the Atlanta Metropolitan Statistical Area (MSA). The results demonstrate that the Machine Learning-enhanced GWR model outperforms traditional methods such as Ordinary Least Squares (OLS) and standard GWR by offering improved accuracy and better insights into spatial heterogeneity. This study contributes to the field of spatial analysis by providing a more comprehensive and interpretable model that can be adapted for various urban phenomena. The findings can enhance our understanding of complex spatial dynamics, support urban planning and public health initiatives, and help policy development by enabling data-driven decisions at the local level.