考虑非均衡性的城市自行车事故骑行者伤害程度影响因素及异质性分析

打开文本图片集

关键词：交通安全；自行车事故；伤害程度；潜在类别分析;Bayes网络中图分类号：U491.31 文献标志码：A文章编号：1002-4026（2026）01-0088-12开放科学（资源服务）标志码（OSID）

Analysis of factors influencing cyclist injury severity and heterogeneity analysis

inurban bicycle accidents considering data imbalance

WANG Chaojian1²，XU Xiaojin'，FENG Bin1，YU Songlin'，ZHANG Weidong'

（1. School of Engineering and Technology， Sichuan Sanhe Colege of Professionals，Luzhou ，China; 2.Luzhou CityResearch CenterforIntelligent Electromechanical ControlEngineeringTechnology，Luzhou ，China）

Abstract：Toexplore the factors influencing the injuryseverityofcyclists inurban bicycleaccidentsandmitigate the impact of data heterogeneityandimbalanceonthequantificationof these factors，this studyproposesamethod integrating resampling，latent class analysis（LCA），and Bayesian networks （BNs）based on 3895 bicycle accidents from the CRSs database.First，LCAwasused to reclassify aident datainto several sub-accident clusters with intra-cluster homogeneity and inter-cluster heterogeneitytoreduce the impact of data heterogeneity.Second，randomover-sampling（ROS）， syntheticminority oversamplingtechnique，andadaptive syntheticsampling approach wereused to resampleeach accident cluster toreduce the impactof data imbalance.Finaly，based on various resampledaccident clusters，two BN structure learningalgorithmsandone parameter learningalgorithm wereappliedandthe optimal BNmodel foreach accidentclusterwasselected basedonAUCvalues toenablequantitativeand heterogeneityanalysesoffactors influencing theinjuryseverityof cyclists.Resultsshowthat when theoverallaccidentdata were dividedintothreehomogeneous sub-clusters，the LCA model achievedan increased entropy valueof 0.943.For the C1，C2，C3，and ODaccident clusters，10，13，9，and12key factorsinfluencing theinjuryseverityof cyclists were identified，respectively.The introduction of LCAand resampling into theBNconsiderablyimproved the BN model’s G-mean value，AUC value，and riskfactoridentficationcapabilies.Factorssuchastimeperiod，cyclist’sgender，cyclist’sage，and weatherconditions showed substantial heterogeneity across different accident clusters.

Key words ： traffc safety； bicycle accidents; injury severity； latent class analysis； Bayesian networks

自行车出行低碳环保，具有良好的社会经济效益和个人健康效益。（剩余15384字）