Development of a Hierarchical Tree Based Regression Model For Rural Traffic Crashes

Abstract

Development of a hierarchical tree based regression model (HTBRM) to find the causal factors of traffic crashes at rural sections of the road is presentenced in this paper. CHAID, a type of rarely used type of HTBRM, was used to model traffic crashes in six national roads of Hokkaido, Japan from 1997 to 2001. CHAID is a non-binary HTBRM; therefore, able to represent the complex relationship between traffic crashes and the independent variables. Here four CHAID models; 1 km non-winter, 1 km winter, 3 km non-winter, and 3 km winter were developed. Models with 3 km road segments were found to be better performing. Tunnels, bridges, and snow sheds were found to be significantly increasing the traffic crashes in winter months of the year. Models lustrated a complex relationship between traffic crashes and the independent variables. Further, several combinations of variables with high number of traffic crashes were identified.