Ridesplitting, a form of ridesourcing services that matches riders with similar routes to the same driver, is a high occupancy travel mode that can bring considerable benefits. However, the current ratio of ridesplitting in the ridesourcing services is relatively low and its influencing factors remain unrevealed. Therefore, this paper uses a machine learning method, gradient boosting decision tree (GBDT) model, to explore the nonlinear effects of built environment on the ridesplitting ratio of origin–destination pairs (census tract to census tract). The GBDT model also provides the relative importance ranking of all the built environment factors. The results indicate that distance to city center, land use diversity and road density are the key influencing factors of ridesplitting ratio. In addition, the non-linear thresholds of built environment factors are identified based on partial dependence plots, which could provide policy implications for the government and transportation network companies to promote ridesplitting.

拼车服务是一种拼车服务形式，可将路线相似的车手与同一个驾驶员相匹配，是一种高占用率出行方式，可带来可观的收益。然而，目前在拼车服务中拼车的比例相对较低，其影响因素尚未得到揭示。因此，本文使用一种机器学习方法，即梯度增强决策树（GBDT）模型，来探讨建筑环境对原点-目的地对（人口普查与人口普查）的乘积比的非线性影响。GBDT模型还提供了所有已构建环境因素的相对重要性等级。结果表明，距市中心的距离，土地利用的多样性和道路密度是影响分乘比的关键因素。此外，