Crash prediction model for basic freeway segments incorporating influence of road geometrics and traffic signs

Dividing a freeway into segments is a fundamental step in establishing its crash prediction model. Instead of using the common segmentation criteria that defines short segments as homogeneous as possible, this study used basic freeway segments that contain heterogeneous geometric and operational characteristics for crash modeling. Variables as cumulative curvature (CUR), cumulative longitudinal gradient (ICUM), side clearance (SideC), and density of traffic signs (DenSig) were proposed to accommodate the possible heterogeneity in these characteristics. The generalized estimating equations (GEEs) were used to model the yearly crash counts (2009-2012) on freeways in Liaoning, China. The modeling results showed that a GEE with autoregressive correlation structure was the best. Accordingly, the overall crash prediction model for all samples and two separate crash prediction models for a two-way four-lane subset and greater than four-lane subset were developed. From these models it could be found that explanatory variables have significant effects on crash counts except for the ICUM. It was also found that the increase in segment length or annual average daily traffic (AADT) could increase the number of crashes, while setting more gradual horizontal curves or widening side clearance could reduce the risk of crash occurrence. In addition, installing more traffic signs within a reasonable density range could lower the crash frequency. This study proposes a new perspective for freeway segmentation and variable preparation that can benefit the road safety practitioner. Meanwhile, analyzing the influence of uncommon variables such as the density of traffic signs on crash occurrence can also provide more insights into the cause of crashes.