A novel approach for signalized intersection crash classification and prediction

This study contributes to the area of safety of signalized intersections by identifying the geometric characteristics that affect the different types of crashes. The frequency of crashes is estimated using the geometric and traffic volume of intersections. Intersections were first classified into two different types: “safe” and “unsafe” based on the average number of crashes by the total number of approach lanes then the frequency of crashes was estimated for each type of intersections separately. This model was then used to estimate the frequency of crashes for a simulation database containing all possible geometric and traffic volume combinations at the intersections, and the output was used to study the effect of the intersection properties on the crash frequency. The number of through lanes on the minor roadway, number of left turning lanes and right turn channelization were among the significant factors in determining the safety of the intersection. In the next part of the study, crashes occurring at signalized intersections were classified into rear-end, angle, turn and sideswipe crash types based on the geometric factors and traffic volume of the intersections and the conditions at the time of the crash. This was achieved by developing another innovative approach we called “Neural Network Trees”. The first neural network model built in the Neural Network tree classified the crashes into either same direction (rear end and sideswipe) or intersecting (angle and turn) crashes. The next models further classified the crashes into their individual types. ADT, number of lanes, number of left turn lanes, right turn channelization, and speed limits were among the significant factors classifying the different type of crashes at intersections. Upon using the models on a simulation database, a greater insight was obtained on the occurrence of various crash types. Multi Layer Perceptron (MLP), Probabilistic Neural Networks (PNN) and Generalized Regression Neural Networks (GRNN) were investigated and used in this study as promising techniques in safety research. The study points to the superiority of estimating traffic crashes by safety level or type compared to estimating one overall aggregate model.