Abstract: In view of the lack of systematic consideration of built environment factors in the analysis of urban traffic accidents, five dimensions including density, diversity, traffic design, accessibility and public transport proximity were selected to characterize the built environment, and four aspects including individual behavior, road conditions, accident types and natural environment were considered at the same time. The machine learning algorithm was used to establish an analysis model that incorporated built environment factors on the severity of urban road traffic accidents. Finally, an empirical analysis was carried out by use of the accident information in a certain district of Chongqing. The research results show that the built environment variables have a greater impact on the severity of accidents. From the perspective of variable importance ranking, the land use mixing degree (14.29%), the density of expressways and trunk roads (12.43%), the density of secondary roads and branch roads (11.54%), population density (11.35%) and accessibility (10.96%) have a higher impact and the cumulative importance reaches 60.57%. At the same time, each variable has a non-linear relationship with the severity of the accident.

Key words: traffic engineering; urban traffic safety; non-linearity; gradient boosting decision tree (GBDT) model; built environment; accident severity

摘要： 针对城市交通事故分析中缺少建成环境因素的系统考虑，以密度、多样性、交通设计、可达性及公共交通临近度等5个维度表征建成环境，同时考虑个体行为、道路情况、事故类型及自然环境4个方面，运用机器学习算法建立融入建成环境因素对城市道路交通事故严重程度影响分析模型；并以重庆市某区的事故数据进行实证分析。研究结果表明：建成环境变量对事故严重程度有较大影响；从变量重要度排序来看，土地利用混合度(14.29%)、快速路及主干路密度(12.43%)、次干路及支路密度(11.54%)、人口密度(11.35%)与可达性(10.96%)的影响程度较高，累计重要度达60.57%；同时各变量与事故严重程度呈现出非线性关系。

关键词: 交通工程；城市交通安全；非线性；梯度提升决策树(GBDT)模型；建成环境；事故严重程度