Examining nonlinear causal relationship between the built environment and VKT using RF–XGBoost

Abstract

Although numerous studies examine the association between the built environment and travel behavior, few carry causal explanations. Using the data from a natural experiment in Shanghai, this study examines the nonlinear causal relationship between the built environment and driving behavior (i.e., vehicle kilometers traveled, VKT) using a hybrid machine learning model that integrates Random Forest (RF) and eXtreme Gradient Boosting (XGBoost). Empirical findings show that the built environment dominantly affects VKT, exhibiting a saliently nonlinear pattern with effective range and threshold. The findings equip policymakers and planners with actionable insight and support for formulating sophisticated transportation intervention strategies to mitigate car dependency. Overall, this study effectively addresses the residential self-selection issue while handling the common multicollinearity trouble among the explanatory variables, providing more accurate causal estimates of the built environment's effect on VKT for nuanced evidence-based guidance in policy making and planning practices.