Trafik Kazaları Analizi için Bayes Ağları Modeli

― The use of high ways as the major means of transportation in Turkey causes a rapid increase in traffic intensity. As a result of the fact that the current infrastructure is unable to respond this rapid increase of traffic intensity, in addition to the traffic infringements made both by drivers and pedestrians, each year a huge number of traffic accidents occur. To prevent the traffic accidents with tangible and intangible losses resulting from it, and to take the necessary precautions in that purpose, it is necessary to conduct a detailed analysis of traffic accidents and the factors influencing its happening. In this research, traffic accidents and the factors influencing traffic accident occurrences are analyzed via Bayesian networks. As a graphical model, Bayesian networks possess a special importance with its abilities such as showing the conditional dependencies between the variables, not being limited to only one output variable, the ability to update the network through evidence observation and the capability to transfer all these information through a graphical interface. In this research, using the official traffic accident reports obtained from Silivri Regional Branch Office and County Gendarmerie Traffic Command a data set is constructed and the corresponding Bayesian network is learned from this data set. Prediction capability of the network is verified through the test data set and the efficiency of the learned model is confirmed with the lift over marginal resulting as positive. Sensitivity analysis is performed for the variables in the network. The proposed model in this research is an exemplary model to analyze the dependency structure between the effects, causes and outcomes of traffic accidents.