A robust spatial modeling framework for walkability capitalization: Multi-weighting analysis with GWR, MGWR, and GNNWR

This study investigates the robustness of building-level walkability measurement and its capitalization into urban land values. A walkability index was constructed from 17 microscale indicators representing accessibility, density, land use, network configuration, and topography. To evaluate methodological robustness, three objective weighting methods (PCA, Entropy, and CRITIC) were compared using correlation analysis and Monte Carlo simulations. The resulting indices were incorporated into OLS, GWR, MGWR, and GNNWR models to examine the spatial heterogeneity of walkability capitalization in the Seoul Station regeneration area. Despite substantial differences in weighting structures, the three methods produced broadly consistent building-level walkability patterns, while empirically derived weighting schemes substantially reduced ranking uncertainty relative to unrestricted weighting scenarios. Among the spatial models, the PCA-based GNNWR achieved the highest predictive performance and revealed pronounced spatial variation in walkability premiums, with stronger capitalization in highly accessible and well-connected neighborhoods. The findings demonstrate that the reliability of walkability measurement depends less on identifying a single optimal weighting scheme than on preserving the underlying covariance structure of urban environmental indicators, providing a robust foundation for spatial analyses of walkability capitalization.