Community park visits determined by the interactions between built environment attributes: An explainable machine learning method

Uncovering the association between built environment (BE) attributes and community park visits by considering potential nonlinear effects can inform more effective spatial policies. This study utilizes real-time population visitation big data to depict the spatial variances in community park visits in the case city of Shenzhen. An explainable machine learning method incorporating random forest and Shapley Additive exPlanations (SHAP) is applied to reveal the relative importance of BE attributes and to examine the nonlinear associations and interaction effects on park visits. The results confirm the decisive roles of park size and walking-based street connectivity on associating with visits, with threshold points at 2 hm2for park size and 0.3 for network warp. The revealed interaction between park size and surrounding BE attributes benefits defining the optimal scale by considering surrounding attributes of both attraction and demand factors. Based on the findings, we further discuss the possible patterns of threshold effects and interaction effects rooted in the examined nonlinearity. The findings guide policy makers in adopting smarter and more effective strategies to improve community park visits. •Visits to 354 community parks were calculated using real-time LBS population estimation.•Random forest (RF) and SHapley Additive exPlanations (SHAP) were used to detect nonlinear and interaction effects.•The indicator measuring walk-based connectivity within a 15-min walking distance is the most important factor besides park size.•Effective ranges of built environment attributes suggest appropriate intervention strategies.•Different types of nonlinear and interaction effects and their generation paths were discussed.