Study On Pedestrian Traffic Accidents Based On Random Forest

As is well known that walking is beneficial to reduce traffic congestion,improve urban traffic efficiency,reduce traffic noise,pollution and greenhouse gas emissions.So it is an economically feasible means of transportation that always be encouraged.However,traffic congestion and many pedestrians disobey rules makes the pedestrian safety a really noticeable problem.The pedestrian traffic safety research is urgently needed.In this paper,the basic theory of pedestrian traffic accident is introduced.Take accident as the research object,the accident severity is assessed from the level and the equivalent of the accident,and the equivalent equation is proposed.The influence factor of pedestrian traffic accident is analyzed form five aspects: pedestrian,vehicle,road,environment and travel time.Secondly,the decision tree is introduced in detail which is the basic theory of random forest model.The sampling method,definition,property and performance indicators of random forest model were explained.Based on the 1000 pedestrian traffic accidents in H city between 2008~2010 and the geometrical characteristics of the intelligent network intersections,the prediction model of pedestrian accident severity based on RF was used to classify the severity of accidents,and compared with the accuracy of models constructed by Bernoulli Naive Bayes,KNN,DT and SVM.Then the importance of the pedestrian characteristic variables is sorted.Delete the less important characteristics will speed things up and meanwhile will have little influence on the accuracy.Lastly,the experiment is designed.The factors of lane-width,period time and the number of conflict positions which are relatively more important and more controllable in pedestrian characteristic variables were analyzed respectively with the RF model.Then the relationships between them and the pedestrian accident severity is figured out.Based on the RF model,a new predictive classification model in pedestrian injury severity is proposed,the new model has better accuracy and stability,less fluctuation compared to the other four models.The study of pedestrian characteristic variables based on the RF model is beneficial to reduce casualties and provide reference for reasonable and safe traffic design and good traffic management.