Non-Linear Associations Between the Urban Built Environment and Commuting Modal Split: A Random Forest Approach and SHAP Evaluation

The study of commuting mode choice is crucial since driving, with all its associated environmental and economic consequences, the United States’ most popular transportation due to urban sprawl, priority road construction America’s love affair automobile. More attention needs be paid sustainable modes such as public transit walking. built environment expected have an impact on choice. Built environments higher density, diversity, intentional design, destination accessibility, shorter distance (collectively known 5 Ds environment) are hypothesized lead more choices, including In this paper, we evaluate variables modal split, four transit-bus, transit-rail, walking, driving. conducted in Mecklenburg County, North Carolina, at geographic level census block groups year 2015. Given complexity relationships environment-travel behavior subject, random forest method used predict aggregated Random employed it capable capturing nonlinear not constrained by limitations other widely methods, multinomial logistic regression. After predicting shares, SHAP values (SHapley Additive exPlanations) choices. As advanced machine learning method, adds explainability model. This resolves limitation methods being “black boxes” converts them “white providing interpretability. They provide insights into both direction magnitude relationships. Thanks rigorous ML-based our helps solidify state knowledge strong evidence that degrees 5Ds choices walking modes. We discuss policy implications study.