Density, diversity, and design: Interpretable machine learning insights into walking distance to transit in Bangkok, Thailand

• Interpretable machine learning uncovers key factors in metro walking access. • Built environment factors outweigh demographics and trips in predicting walk distance. • Compact, diverse, and connected station areas shorten walking access in Bangkok. • Threshold and interaction effects explain how density, diversity, and design interact. • Findings guide transit-oriented planning for walkable, equitable metro access. Enhancing walking accessibility to metro stations is a critical element of transit-oriented development (TOD), yet the combined influence of built environment, demographic, and trip factors remains insufficiently understood. This study employs interpretable machine learning—XGBoost with SHapley Additive exPlanations (SHAP)—to examine nonlinear and interaction effects shaping walking distances to metro stations in Bangkok, Thailand. Drawing on detailed walking route data from 1500 passengers across 50 stations, the analysis identifies built environment characteristics as the most influential predictors, surpassing demographic and trip-related factors. Results reveal important threshold and interaction effects: compact, high-density station areas with fine-grained street networks consistently shorten walking distances, while dispersed points of interest and disconnected green spaces can lengthen them, even in otherwise well-developed areas. These findings reinforce the TOD “3Ds” framework—Density, Diversity, and Design—by highlighting the need for integrated land-use planning, connected pedestrian infrastructure, and purposeful green space integration. The study offers actionable, context-specific guidance for improving walkability and equitable metro access in Bangkok and other rapidly urbanizing regions.