摘要
为降低国省道机动车交通事故严重程度,提取出国省道机动车死亡交通事故的显著特征,提出1种基于SMOTE-CatBoost模型的交通事故显著特征研究方法,对2016—2020年N市国省道机动车交通事故特征进行研究。首先,根据基础数据集的不平衡分布情况采用SMOTE平衡算法形成平衡数据集以训练机器学习模型;其次,选用CatBoost模型构建国省道机动车交通事故预测模型;最后,采用SHAP可解释性算法挖掘国省道机动车死亡交通事故各个特征与事故的关联程度。结果表明:CatBoost模型的综合性能优于其余4种机器学习模型,其准确率、召回率、F1分数、AUC值分别比排名第2模型高8.5%、7.45%、7.09%、7.47%;死亡交通事故中显著特征有驾驶人在驾车时有其他妨碍安全驾驶行为、机动车与行人的碰撞等6种。本研究成果可为公安工作以及道路交通管理部门人员提供国省道道路死亡交通事故的显著特征,为减少死亡交通事故发生以及改善国省道道路行车环境提供一定的理论支撑。
To reduce the severity of motor vehicle traffic accidents on national and provincial highways and extract the salient features of motor vehicle death traffic accidents on national and provincial highways,this paper proposes a method for studying the salient features of traffic accidents based on the SMOTE-CatBoost model and studies the characteristics of motor vehicle traffic accidents on national and provincial highways in N city from 2016 to 2020.First,according to the unbalanced distribution of the basic data set,the SMOTE balancing algorithm forms a balanced data set to train the machine learning model.Secondly,the CatBoost model is used to build a national and provincial highway motor vehicle traffic accident prediction model.Finally,the SHAP interpretability algorithm is used to explore the correlation between various characteristics of motor vehicle traffic accidents on national and provincial highways and the accident.The results show that the overall performance of the CatBoost model is better than the other four machine learning models,and its accuracy,recall rate,F1 score,and AUC value are 8.5%,7.45%,7.09%,and 7.47%higher than the second-ranked model respectively;The salient features of traffic accidents include six types of behaviors that others behaviors that hinder safe driving and collisions between motor vehicles and pedestrians.The results of this study can provide public security and road traffic management department personnel with the salient characteristics of serious traffic accidents on national and provincial highways,and provide certain theoretical support for reducing fatal traffic accidents and improving the driving environment on national and provincial highways.
作者
王誉翔
马社强
赵丹
王晟由
WANG Yuxiang;MA Sheqiang;ZHAO Dan;WANG Shengyou(School of Traffic Management,People's Public Security University of China,Beijing 100038,China)
出处
《交通工程》
2024年第12期40-46,共7页
Journal of Transportation Engineering
基金
国家重点研发计划“重大自然灾害防控与公共安全”专项(2023YFC3009702)。
作者简介
王誉翔(2001-),男,硕士研究生,研究方向为交通安全。E-mail:794388997@qq.com;通讯作者:马社强(1973-),男,博士研究生,副教授,研究方向为交通安全。E-mail:masheqiang@163.com。
