Development of Zonal Safety Performance Functions for Local Road Intersections and Segments in Connecticut

Crash counts at a location are predicted using a Safety Performance Function (SPF) equation. The expected number of crashes on a traffic facility can be estimated using SPFs and the required countermeasures can be taken to reduce crashes in future. Due to the absence of sufficient traffic count data for local roads, new crash prediction approaches are essential to implementing highway safety improvement strategies. The study focuses on developing SPFs with coefficients varying by geographic covariate for segment and intersection crashes on local roads in Connecticut. Demographic and network topology data has been used as a surrogate for traffic count data which are not available for these roads. Two clustering methods – K-Means and Latent Class Clustering (LCC) has been explored for classifying cases for varying coefficients. The variables that were used to classify into clusters were land cover, population density and employment density. The models clustered using LCC with total population, retail and non-retail employment and average household income as independent variables were found to be the best based on model fit and out of sample prediction.