Development of a Segmentation Method Considering Crash Frequency Criteria

Abstract

Recently, road safety has been an important issue around the world in an effort to reduce injuries and deaths that occur due to vehicle crashes. Many research related to predicting crash frequency have been done and out of the existing crash prediction models, the safety performance function is the most widely used. Safety performance functions are developed using regression analysis and in order to do this, the crash data that is going to be used for the analysis must be divided via segmentation from a raw dataset so that the independent variables and dependent variables can be determined. Segmentation is one of the most important factor for developing SPF because the resulting SPF and its performance to predict crash frequency vary on the segmentation method used. This means that the success in development of an safety performance function for roadway depends on how well the data can be organized into distinct homogeneous entities or on the approach of segmentation. Currently, many different segmentation methods are being used for safety performance function research but there is no clear guideline for selecting proper and suitable segmentation method to be used in estimating SPFs. Initially, this research focuses on applying the existing method onto Korean freeway crash data. As a result, no apparent favorite method is determined due to the advantage and disadvantage each method possesses. To this end, a new segmentation method, where two of the former existing methods are applied to one single freeway line is developed. The dataset is first segmented using the "homogeneous AADT" method and then the segments that show low crash predictability are selected to be disaggregated into smaller segments using the "homogeneous curvature" method. By doing all this, the advantages of both the segmentation method can be applied and also the crash predictability of the estimated safety performance function will increase due to the segments with high error being eliminated. As a result of using this new method, the crash predictability of the safety performance function of all 8 Korean freeway lines significantly increased to that of using former methods.