Neighborhood-Scale Built Environment and Low-Carbon Travel: Empirical Evidence from 108 Communities in Beijing’s Fengtai District

While the linear and homogeneous effects of the built environment on travel behavior have been widely examined, the nonlinear patterns and contextual heterogeneity remain insufficiently explored—particularly in Chinese megacity suburbs. Recent methodological advances, such as double machine learning and gradient boosting decision trees, have begun to reveal threshold effects and non-linear moderating mechanisms that conventional linear models cannot capture. This study examines the associations between neighborhood-scale built environment characteristics and residents’ travel-related carbon emissions, based on 108 communities in Beijing’s Fengtai District, China. Using a “5D+3G” indicator system and combining hierarchical regression, subgroup regression, and random forest, the analysis identifies exploratory nonlinear patterns in the built environment–emissions relationship. The random forest model showed notable overfitting given the modest sample size (training R2 = 0.730; test R2 = 0.223; cross-validation mean = −0.172), and the identified thresholds are therefore treated as hypothesis-generating rather than confirmatory. Linear models reveal that most built-environment attributes exhibit weak and statistically insignificant associations with emissions, indicating that average-effect specifications may obscure the complexity of these relationships. Rail-transit accessibility shows distance-based attenuation: emissions remain low within the exploratory range of approximately 500 m of metro stations, increase between 500 and 1500 m, and plateau beyond 1500 m in this sample. POI functional density exhibits diminishing marginal returns, with a sample-specific saturation pattern near 4 POIs/ha. The built environment–emissions relationship thus appears more heterogeneous and context-dependent than previously assumed. Based on these empirical patterns, the study proposes a five-category neighborhood typology and corresponding design intervention strategies tailored to local conditions. The analytical framework offers a transferable reference for architects, urban designers, and planners seeking evidence-based guidance for low-carbon neighborhood design in similar contexts, while the specific thresholds require validation in other settings.