Unraveling nonlinear and interaction effects of multilevel built environment features on outdoor jogging with explainable machine learning

Built environment (BE) is widely perceived as crucial in promoting physical activities (PA). However, the nonlinear and interaction relationships between multilevel (macroscale and microscale) BE and outdoor jogging are poorly explained. Therefore, this work established an explainable machine learning-based analytical framework that integrating Gradient Boosting Decision Tree and SHapley Additive exPlanations to interpret the nonlinear and interaction effects. Employing Beijing, China as the case, this study conducted an empirical analysis of multilevel BE factors affecting outdoor jogging utilizing multi-source big data, including large-scale jogging GPS trajectories and street view images. Results revealed that: (1) Macroscale BE including sports amenities and accessibility play an important role in affecting jogging. (2) All BE factors have nonlinear association with jogging, and the effective range and threshold effects vary by variables. Notably, several variables associated with jogging are in inverted U or V shapes. (3) The complex interaction effects including synergistic, weakened, among BE factors were scrutinized. BE factors regarding accessibility and sports amenity interact more easily. (4) Regions with similar local effects were clustered to understand the spatial varying of BE's influences and establish targeted interventions. These findings can help urban planners design more nuanced intervention strategies for supporting outdoor jogging.