自行车驾驶员交通行为方式判定研究

Traffic Manner Judgment for Bicyclist

鉴于我国交通事故中涉案者交通行为方式对于事故责任认定具有重要作用,提出一种基于特征选择的自行车驾驶员交通行为方式判定方法。采集交通事故案例数据并构建组合特征后,基于SMOTE算法进行数据不平衡处理。针对不同数据集构建支持向量机、随机森林和人工神经网络这3种多分类模型,运用交叉验证和受试者工作特征曲线进行模型评价,确定最优分类准确率和关键特征变量。研究表明:该判定方法准确率在组合数据集可达78.83%,在原始采样数据集可达82.19%;车座损伤、车座旋转、机动车类型、车把旋转是交通行为方式判定的关键特征变量。

In view that the traffic manner of the involved persons in the traffic accidents plays an important role for responsibility identification in China,a method of judging the traffic manner of bicyclists based on feature selection was proposed.After collecting traffic accident case data and constructing combination features,the data imbalance processing was carried out on the basis of SMOTE algorithm.The three kinds of multi-classification models,that is support vector machine,random forest and artificial neural network,were built according to different data sets and evaluated by the way of cross validation and receiver operating characteristic curve.The optimal classification accuracy and key characteristic variables were ascertained.The results show that the identification accuracy rate of the proposed method is 78.83%in the combined data sets and 82.19%in the original sampling data sets.Bicycle saddle damage,bicycle saddle rotation,vehicle type and bicycle handlebar rotation are the key characteristic variables for traffic manner judgement.