Development of Crash Frequency Prediction Model and Identification of Hazardous site locations: A case study of BP highway

Abstract

Crash prediction models (CPMs) have been used in many countries as a useful tool for road safety analysis and design. Each model is different in terms of methodology, data accuracy, variability in highway geometry and predictor variables used to predict crashes. This research focuses on developing a relationship between crash counts and roadway attributes, namely curve density, length of horizontal curves, maximum length of continuous tangent, maximum longitudinal grade, average longitudinal grade, access density, minimum sight distance within a segment, minimum radius of curvature and average lane width. Generalized Linear Modelling Technique based on Poisson distribution was selected for the development of model. The model was developed using the crash and road attribute data of Section II of BP highway. Out of the predictor variables, access density, minimum horizontal sight distance, maximum length of continuous tangent and minimum radius of curvature were found to be the most significant predictors. The proposed model was validated using crash and road attribute data from Section III of BP highway. The R2 values obtained for the initial developed model was 0.509 whereas the one obtained during model validation was 0.4308. R2 value obtained for the final model using both the core data-set and the data used for validation was obtained as 0.516.