Using data mining techniques in clustering and identify patterns of road accidents

Clustering of road accident data by using data mining algorithms lead to derive a set of rules that can be used by traffic police and organizations such as transportation terminals and highway engineers to improve road safety. In this study to implement data mining techniques, Standard Model CRISP- DM is used. The required data includes 14960 accidents of Karaj- Chaloos road during 90- 93 year. For clustering, K-means, Kohonen and Two Step algorithms is used. In order to ensure data correlation with the severity of road accidents we use Apriori correlation model and finally to analyze the severity of accidents on roads classification and regression tree model (C & R tree) is used. The results of the clustering model indicates that the variables contributing in road accidentincluding geometry, accident location, direction of the way, lining the road, the shoulder of the road, road surface conditions, location and type of repairs procedures respectively. The results of the correlation model indicate that among tree accident type (injury, death and damage) the most percentage of accident type is regarded as damage.Classification and Regression Tree algorithm. Results show the percentage degree of each variable that lead to three accident type that the most important is existence of vision obstruct.