基于集成学习的交通事故严重程度预测

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2025.16.011

中图分类号：TN912-34；U491.3 文献标识码：A 文章编号：1004-373X（2025）16-0061-06

Traffic accident severity prediction based on ensemble learning

JIA Xianguang'，SONG Tengfei'，LU Yingying² （1.SchoolofTransportationEngineering，KunmingUniversityof Technology，Kunming 65o5oo，China; 2.SchoolofIformationEngineeringndAutomation，KunmingUiversityofTechoogyKunming65O5，a）

Abstract：Inordertoimprovetheperformanceofroad traffcaccidentseverityprediction modelsandanalyze theimpactof acidentfeaturesonacidentseverityamethodoftraffcaccidentseveritypredictionbasedonadouble-layerStackingodelis proposed.The BSMOTE2 algorithm isused tobalancethedataandverifywhetherdatabalancing procesing willhaveapositive impact on model prediction.The GBDT-RFECV algorithm isused for k -fold cross validation selection to complete the feature dimensionalityreduction.Atwo-layer Stacking model isbuilt.Thefirstlayeriscomposedof BiGRUandXGBoost，using time seriesfeatures forBiGRUandstaticfeaturesforXGBostforthepreliminaryprediction.TheCatBoostmodelisusedatthe secondlayerandcombinedwith thepredictionresultsofthefirstlayerforthefinalseverityprediction.Theresearchresults indicate that theaccuracyofthemodel，macro F1 ，andmacroAUChaveallimproved significantly，indicatingthatdatabalance processing hasapositiveimpactonmodelprediction.IncomparisonwithKNN，BiGRU，RF，andXGBoost models，theproposed double-layer Stacking model can improve prediction accuracy by 5.45%，10.23%， 1.78% ，and 2.34%，respectively，the macro F1 （204 value can be increased by 5.31% ， 9.91% ，1.35%，and 1.92%，respectively，and the macro AUC canbe increased by11.13%， （204 6.97% ， 2.13% ，and 2.71%，respectively.The double-layer Stacking model can perform beter than other modelson multiple evaluation metrics.

Keywords：traficsafety;traffcaccidentseverity；predictiveanalysis;ensemble learing;machine learning;deeplearning; feature dimensionalityreduction

0 引言

随着社会经济的快速发展，汽车保有量逐渐增加，但道路交通安全问题日益突出。（剩余7611字）