Spatiotemporal heterogeneity of the built environment's impact on ridesplitting behavior

Ridesplitting is a ridesourcing service where multiple passengers with similar routes share a single trip. Understanding the factors influencing ridesplitting behavior is vital for sustainable urban transport planning. However, most studies analyze ridesplitting behavior at aggregated spatial units (e.g., census tracts), assume uniform trip attributes within an area and overlook order-level heterogeneity, which can lead to biased results. Thus, based on a real-world DiDi Chuxing dataset, this study employs a gradient boosting decision tree (GBDT) model to examine the nonlinear effects of trip attributes and built environment (BE) factors on ridesplitting behavior, measured by the ridesplitting probability (RP) at the individual/order level. BE factors are measured within a 1-km buffer around trip origins and destinations, and the analysis is conducted separately for weekdays and weekends. The GBDT results reveal that BE variables contribute 61% of the model's predictive performance, with origin-side factors having a slightly stronger impact than those at the destination. Additionally, this study also identifies nonlinear relationships and threshold effects between variables and RP. It is observed that RP exhibits pronounced temporal heterogeneity throughout the day. The order density, trip distance, distance from the trip origin to the city center, and land use mix at the trip origin are impactful only within specific ranges. Subways compete more with ridesplitting than complements it, and subway accessibility exerts a greater influence than bus. The findings of this study provide key insights and actionable recommendations for improving urban transport efficiency and promoting sustainable development.