Analyzing Traffic Accident Situation With Bayesian Network

With the huge progress of China in social, economic, motorization andtransportation recently, the level of road traffic safety is getting lower. Trafficaccidents cause not only casualty and property damage, but also traffic delay and jam,fire, explosion, and leakage of dangerous goods sometimes, which lead to huge lossfor the people. In order to make the social and economic develop rapidly and reducethe damage caused by traffic accident, it is necessary for the government to figure outhow to enhance traffic safety and prevent from traffic accident.The goal of situation analysis is to examine the relationship among trafficaccident and the potential factors, including people, vehicle, road and environment, aswell as to analyze the process of developing and changing of traffic accident. Basedon situation analysis, effective traffic safety management measures can be made sothat the damage caused by traffic accident can be reduced. Therefore, the researchconcerning about situation analysis is of great importance for enhancing traffic safetyand ensuring the rapid social and economic development.Founded by National High-tech R&D Program of China (863Program)-QuickInvestigation and Disposal Technologies of Major Accidents, a forecasting model wasdeveloped for traffic accident situation, and the relationship between factors andtraffic accident was analyzed according to the conditions and features of the trafficsystem and social-economic development in China. It then proposed some trafficmanagement measures and developed a program of traffic accident rapid response toimprove traffic safety.Two potential methods were studied for situation analysis, which are Bayesiannetwork and discrete choice model, and employed them respectively to modelsituation of motor vehicle accident. Bayesian network, which was proved to be betterthan discrete choice model in prediction accuracy, was picked out and employed inmodeling the situations of not only motor vehicle accident, but also non-motorvehicle/pedestrian accident. Then the impacts of the factors on accident situation arelearned by using the Clique Tree Propagation based on the Bayesian networks, andsome traffic management strategies were proposed to reduce damage caused by trafficaccident. A forecasting model of accident duration was also developed according to the theory of survival analysis. Based on the models of situation analysis and durationforecasting, a rapid response system was then developed. The incremental learning ofBayesian network was concerned about at the end of the dissertation.The major contributions of this dissertation are as follows:(1) developing situation analysis models for motor vehicle accident andnon-motor vehicle/pedestrian accident, respectively, with Bayesian network, andanalyzing the relationship among dependent variables and independent variables byusing the methods of both structural description and quantitative deduction;(2) comparing Bayesian network and discrete choice model with respect to thefactors of both qualitative analysis and quantitative calculation, including variableschoosing, relationship analysis and goodness-of-fit of the models, etc.;(3) inferring the impacts of factors on accident situation with the Clique TreePropagation, and proposing and evaluating traffic management measures according tothe inferred results;(4) studying on the incremental learning of Bayesian network, and suggestingthat the structure of the Bayesian network should be learned again if the number ofnew data reach10%of the original data;(5) developing a model for accident duration forecasting based on the theory ofsurvival analysis.The results can be used to analyze the process of developing and changing oftraffic accident under the influence of the factors and to propose effective trafficsafety management measures. It also provides useful models for prediction of accidentsituation and duration as well as developing rapid response system for traffic accident.The study contributes to the improvement of road traffic safety, reducing the casualtyand property damage caused by traffic accident, and making the social and economicdevelop rapidly.