城市道路交通事故形态影响因素分析与预测

Influencing factors analysis and prediction of urban road traffic accident patterns

为探究不同形态城市道路交通事故的发生原因,将事故形态的影响因素进行筛选和约简,选取3种不同算法对事故形态进行分析与预测,对比预测模型的准确性。采用粗糙集理论对原始交通事故形态影响因素变量进行转换和约简,获得满足建模要求的试验数据,并按照总体一致原则把数据等分为训练集和测试集。基于C5.0决策树算法,构建交通事故形态预测决策树模型并进行模型准确性验证,生成交通事故形态规则集;另外,采用似然比检验筛选自变量构建交通事故形态多元Logistic回归预测模型;构建多层感知器(MLP)神经网络预测模型,检验模型训练集与测试集的准确率并进行对比分析。结果表明:3种模型中,C5.0决策树算法对交通事故形态在训练集和测试集中的预测准确率分别为80.39%与79.63%,高于多元Logistic回归模型和MLP神经网络模型。采用C5.0决策树算法得到交通事故形态主要影响因素为交通方式的选取,行驶在道路横断面位置,违法行为与行驶状态等,解释性良好。研究可为分析及预测城市道路交通事故形态,分析事故产生原因提供方法参考,还可为交通管理部门提供决策依据。

In order to explore the causes of urban road traffic accidents,according to different traffic accident forms,the influencing factors of accidents were screened and reduced,and three different algorithms were selected to analyze and compare the accident patterns with the prediction models.The rough set theory was used to transfer and simplify the original traffic accident data,the experimental data that meet the modeling requirements was obtained.The data was equally divided into the training set and test set according to the principle of consistent overall characteristics.Based on the C5.0 decision tree algorithm,the traffic accident prediction model was constructed,and the accuracy of the model was verified,the traffic accident pattern rule set was generated.In addition,the likelihood ratio test was used to screen the independent variables to construct the multiple logistic regression model,and the multilayer perception(MLP)neural network traffic accident pattern prediction model was constructed to test the accuracy of the model training set and test set.The results show that among the three models,the accuracy of C5.0 decision tree model of training set and test set is the highest,80.39%and 79.63%respectively,the accuracy of C5.0 decision tree model in predicting traffic accident pattern is higher than the multivariate Logistic model and MLP neural network model.Through the C5.0 decision tree model,the main influencing factors of traffic accidents patterns were got as selection of traffic mode,driving at the cross-section of the road,illegal behavior and driving status.The research provide a method for predicting the patterns of urban road traffic accidents and analyzing the causes of accidents,and provide decision-making basis for traffic management departments.10 tabs,1 fig,27 refs.