Applying machine learning to decode built environment thresholds for public and active transport distances in the global south

Urban mobility in rapidly growing megacities, particularly in the Global South, presents unique challenges due to population densities, fragmented transit networks, and informal urban growth. While extensive research has examined how the built environment (BE) influences transport mode choice, the impact of BE on active transport (AT) and public transport (PT) trip distances remains partially underexplored, particularly in India. The current study addresses this gap by analyzing the non-linear effects of BE characteristics on trip distances for both AT and PT in the Mumbai Metropolitan Region. By using gradient boosting decision trees, alongside six other machine learning (ML) models, including support vector machines, random forest, and artificial neural networks, this study identifies key BE factors that shape trip distances. The findings reveal that BE variables account for 66 % of the variance in AT distance and 63 % in PT distance. This highlights the dominant role of urban form over socio-demographic factors. Notably, proximity to railway stations, land-use diversity, and intersection density exhibit strong threshold effects on travel distances. Additionally, partial dependence plots uncover non-linear BE-travel behavior interactions demonstrating that moderate densities (300–500 blocks/km²) optimize AT, while PT ridership is more sensitive to network accessibility. The study provides data-driven policy recommendations to enhance pedestrian infrastructure, refine transit-oriented development, and promote sustainable multimodal mobility. By integrating advanced ML methods with transportation policy analysis, this research bridges critical methodological and contextual gaps thus offering actionable insights for urban planners in high-density, transit-dependent cities worldwide.