基于改进决策树的道路交通事故成因耦合分析  

Coupling Analysis of Causes of Road Traffic Accidents Based on Improved Decision Tree

作　　者：郭兰英[1] 张晓静 程鑫[1] GUO Lan-ying;ZHANG Xiao-jing;CHENG Xin(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,陕西西安710064

出　　处：《计算机技术与发展》2022年第6期156-161,166,共7页Computer Technology and Development

基　　金：国家重点研发计划项目(2018YFB1600800);陕西省重点研发计划(2020GY-018);中央高校基本科研业务费专项资金项目(300102241305)。

摘　　要：道路交通事故的发生充满不确定性,为尽可能减轻事故损害,研究分析其成因具有重要意义。该文提出了一种基于改进决策树的道路交通事故成因耦合分析算法。算法采取数据清洗、方差过滤、相关性分析、数据重采样等方法预处理西安市历史事故数据,分析道路交通事故特征中的时间特性及道路环境特性,发现交通事故的严重程度易受交通特征影响。以事故严重程度为目标,初步选取道路因素及环境因素在内的17个交通特征候选自变量,以梯度提升决策树(gradient boosting decision tree,GBDT)为基模型,结合递归特征消除法(recursive feature elimination,RFE)探究不同交通特征对事故等级的影响。利用GBDT-RFE模型筛选出道路类型、道路横断面位置、交通信号方式、路侧防护设施类型、道路线型、照明条件作为主要特征变量,以此构建决策树模型,提取易引发重大事故且置信度较高的多因素耦合模式作为重点情境防范,为事故安全预警提供参考。The occurrence of road traffic accidents is full of uncertainty.In order to reduce the accident damage as much as possible,it is of great significance to study and analyze its causes.We present an algorithm for coupling analysis of the causes of road traffic accidents based on improved decision trees.The algorithm uses data cleaning,variance filtering,correlation analysis,data resampling and other methods to preprocess historical accident data in Xi’an,analyze the time characteristics of road traffic accident characteristics and road environment characteristics,and find that the severity of traffic accidents is easily affected by traffic characteristics.Aiming at the severity of the accident,17 traffic feature candidate independent variables including road factors and environmental factors are initially selected,and the gradient boosting decision tree(GBDT)is used as the base model,combined with the recursive feature elimination method(RFE)explore the impact of different traffic characteristics on the level of accidents.Use the GBDT-RFE model to screen out the road type,road cross-section location,traffic signal mode,roadside protection facility type,road alignment,and lighting conditions as the main feature variables,and build a decision tree model to extract the high-confidence multi-factor coupling mode that is likely to cause major accidents as key situation prevention and provides a reference for accident safety early warning.

关 键 词：交通事故 GBDT-RFE 事故严重程度 决策树 多因素耦合 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]