Study Of Road Traffic Accident Sequence Pattern And Severity Prediction Based On Data Mining

With the continuous increase of the quantity of motor vehicles, the road traffic safety problems have become increasingly prominent. Countries around the world not only pay close attention but also long-term commitment on the measures to reduce the number of road traffic accidents or loss of personnel and property involved. The occurrence of road traffic accidents has close relationship with people, vehicle, road, and environment factors in traffic system. The disordered coupling between the external factor as traffic environmental factors and the internal factor as driver’s behavior factors leads to the occurrence of the accidents. Road traffic accident data directly reflects the interaction relation between each influence factor. Using data mining techniques for in-depth study of the potential rules and accident mechanism provides the theoretical support for the traffic safety management and driving safety education.The key point of this study is the sequence pattern analysis of influence factors of road traffic accidents and accidents severity prediction. This study also discussed the impact model of the factors which strongly influence the accidents. At first, through the analysis of road traffic accidents data characteristics, it was found that the factors involved in the process of the accident form a sequence according to time order and have comprehensive influence on the accidents. In this way, the sequence pattern data mining method is chosen to develop the high frequency sequence pattern of the factors and accident results in two aspects. Moreover, the frequent degree index which considers the proportion and the weight factors and better suits this study is put forward. In addition, in order to qualitatively predict the accident severity represented by injury severity, the prediction model is developed using CHAID decision tree method to study influence factors which easily leads to serious traffic accidents.Traffic environment factors sequential pattern mining found high frequency sequence patterns involving influencing factors such as atmospheric conditions, regional population size, lighting and accident results factors represented by collision manner and maximum injury severity. The sequence patterns revealed the comprehensive influence modes of the influence factors on accident results. Research found that "rain" and "night without lighting" have larger possibility to cause serious accidents. Sequence pattern mining of driver’s behavior factors also found high frequency sequence patterns involving influencing factors such as driver’s age, driver’s sex, behavior prior to accident, special event priori to accident and accident results represented by collision manner and injury severity. The high frequency sequence patterns can considered as qualitative prediction to accident results in given traffic factors situations. The two categories of sequence patterns have good reference value for the analysis of road traffic accidents mechanism and the driving safety education.Two decision tree models are established in two aspects separately to predict the accident severity. It was found that collision manner has biggest influence on the accident severity. Tree models found similar conclusions with the sequence pattern analysis, suggesting that "rain", "nigh", "collide with vehicle in opposite direction", and "lane change collision" are the factors that have larger possibility to lead serious accidents.