Nonlinear effects of built environment on ridesplitting ratio: Discrepancies across sharing motivations

Ridesplitting consolidates passengers with similar routes, offering a sustainable alternative that enhances traffic efficiency, mitigates congestion, and reduces air pollution. However, currently the ridesplitting ratio remains low, with existing research inadequately addressing the combined effects of sharing motivations and built environment on its adoption. To address this gap, we develop a rule-based algorithm to infer ridesplitting motivations based on individual travel patterns, extracted from a massive set of observed ride-hailing data, the thresholds of which are determined by a time-based-sampling validation method. Employing eXtreme Gradient Boosting (XGBoost) models and partial dependence plots (PDP), we further analyze the nonlinear impacts of the built environment on the ridesplitting ratio across different user groups. Results indicate that our algorithm significantly outperforms k-means clustering in terms of accuracy, with company density at the destination being a key determinant of the ridesplitting ratio. Additionally, significant heterogeneity is observed in both travel patterns and the nonlinear effects of the built environment among different user groups. This study provides valuable insights on how governments and transportation network companies (TNCs) could promote ridesplitting services through strategic modifications based on the built environment.