Complex built environment and explicit distance threshold: A machine learning diagnosis of commuting active travel behavior

Active travel is a key pathway to achieving urban transport emission reduction and sustainable development. Existing studies have demonstrated a nonlinear relationship between the built environment and active travel, but the interaction effects between individual attributes and environmental attributes at the micro-scale have not been fully explored. This study draws on large-scale origin-destination (OD) commuting data from internet map services in Shenzhen, China, to develop a LightGBM machine learning model and integrates the SHAP interpretability framework to quantify feature contributions and identify nonlinear thresholds, thereby elucidating the decision logic of active commuting. The results show that: (1) commuting distance is the primary constraint, exhibiting a threshold effect, with active travel peaking at 1.8 km and exhibits a sharp decline beyond 7.38 km; (2) built environment effects display spatial asymmetry, with residential environments exerting stronger influence than workplace environments—higher residential POI and population density promote active travel, whereas high-intensity workplace development without adequate support may inhibit it; (3) socioeconomic factors regulate behavioral elasticity through distance-dependent effects: health awareness dominates in short-distance contexts, whereas motorization capacity (income and car ownership) prevails over longer distances, exacerbating travel inequality. These findings provide empirical support for spatially differentiated active travel policies and the targeted allocation of non-motorized transport infrastructure in high-density, compact urban contexts.