Study of Signalized Intersection Crashes Using Artificial Intelligence Methods

High percentage of traffic crashes occurred at intersections. Generally, human error is the only cause to be blamed. However, approaching roadside environment that drivers confronted with right before crash occurrence is actually critical to crash occurrence. In this study, environmental factors critical to intersection crashes occurrence were identified via negative binomial regression and artificial neural networks. With these factors, data mining was then applied to Find the rule for judging intersections safety. The 3,441 crashes occurred at 102 intersections in Taichung, Taiwan during 1999 similar to 2004 were collected. Numbers of crashes in specific approaching direction combinations were then modeled. It was found that geometry of approaching roadways was indeed critical. Total 47 safety rules generated from Gini decision tree can serve as a tool for safety evaluation of intersections. Finally, although no single factor can induce crashes alone, road width seems to be a crucial factor for intersection-related crash occurrence.