GALAX: A Framework for Geospatial Analysis Leveraging AutoML and eXplainable AI

We introduce GALAX (Geospatial Analysis Leveraging AutoML and eXplainable AI), an enhanced geographically weighted regression (GWR) framework integrating automated machine learning (AutoML) and explainable artificial intelligence (XAI). GALAX employs AutoML to automatically optimize the best machine learning models for different geographic regions and XAI, via SHapley Additive Explanations, to provide localized interpretability through systematic quantification of feature contributions. GALAX can capture nonstationary and nonlinear relationships, analyze high-dimensional feature spaces, and support both regression and classification applications. We validate GALAX through mobility pattern analyses within the Philadelphia Metropolitan Statistical Area. For regression, GALAX outperforms GWR and geographical random forests (GRF) with root mean square error reductions of 36. 0 percent and 44. 0 percent, and R2 improvements of 21. 5 percent and 38. 2 percent, respectively. GALAX reveals nonlinear relationships and spatial heterogeneity in factor importance, with median household income showing generally positive correlation with mobility distance, with a more moderate slope above 150, 000, and being the dominant factor in suburban areas. For classification, GALAX achieves higher accuracy than GRF, uncovering spatial variations, nonlinear relationships between factors and mobility categories, and specific threshold effects missed by linear models. GALAX bridges machine learning and traditional geographical methods, offering new possibilities for analyzing spatial phenomena while maintaining interpretability.