Identifying Effective Factors and Predicting Road Accidents Using Data Mining Approaches (Case Study of Tehran-Qom)

The main purpose of this study, Modeling, identifying key factors and investigating various data mining algorithms in traffic accidents in Iran, especially on urban roads. The following was extracted set of rules that can be effective in identifying key factors and their impact in reducing accidents. The data included 5099 records of traffic accidents on the Tehran-Qom road in Tehran during a four-year period. Each accident record consisted of 57 accidents related to accidents. In the pre-processing, the greedy search algorithm was used to find the best sub-set of factors. After the pre-processing, reduced records used. Finally, 8 characteristics, accident factor, type of collision, visual impedance, accident situation, road surface conditions, location geometry, direct causes and type of accident were investigated. To achieve the objectives of this research, seven different data mining techniques were used. 6 Data mining algorithms were used to predict and identify key factors using WEKA data mining software; These algorithms include: J48, PART, Logistic, Classification via Regression, Naïve Bayesian and Multilayer Perceptron. Also, the apriori algorithm was used to extract the rules. The results showed that Perceptron and Part algorithms had the best performance among other algorithms to predict. Algorithm Naïve Bayesian despite being in the range of acceptable has a weaker performance than other algorithms, both in model accuracy and coverage. Several of the most important factors of the set of rules were road accidents, flat and straight geometry, Lack of visibility barrier And attending the accident site.