DEVELOPING CRASH PREDICTIVE MODELS FOR A PRINCIPAL ARTERIAL

Two techniques of modeling crash occurrence were investigated. A principal arterial in Central Florida was selected to calibrate the models. The modeling effect involved using both the multiple linear regression and the Poisson regression methodologies. Results illustrate that the Poisson regression approach is superior to linear regression. The results also show the significance of the Annual Average Daily Traffic (AADT), and the log of the section's length on the frequency of crashes. Several geometric design variables also affect crash occurrence, including the degree of horizontal curvature, shoulder, median, and lane widths, and whether the location of the crash was urban or rural. These results have important implications for developing predictive models of crash occurrence and in understanding the factors that influence them.