Single-vehicle Run-off-road Crash Prediction Model Associated with Pavement Characteristics

This study aims to evaluate the impact of pavement physical characteristics on the frequency of single-vehicle run-off-road (ROR) crashes in two-lane separated rural highways. In order to achieve this goal and to introduce the most accurate crash prediction model (CPM), authors have tried to develop generalized linear models, including the Poisson regression (PR), negative binomial regression (NBR)), and non-linear negative binomial regression models. Besides exposure parameters, the examined pavement physical characteristics explanatory variables contain pavement condition index (PCI), international roughness index (IRI) and ride number (RN). The forward procedure was conducted by which the variables were added to the core model one by one. In the non-linear procedure and at each step, 39 functional forms were checked to see whether the new model gives better fitness than the core/previous model. Several measurements were taken to assess the fitness of the model. In addition, other measurements were employed to estimate an external model validation and an error structure. Results showed that in PR and NBR models, variables coefficients were not significant. Findings of the suggested nonlinear model confirmed that PCI, as an objective variable, follows the experts anticipation (i.e., better pavement manner associates with less ROR crashes). Finally, it should be noted that the roughness variable was insignificant at the assumed significance level, so it had no contribution to ROR crashes. The results imply that improving the pavement condition leads to a more probable decrease in the ROR crashes frequency.

Keywords:

Highway Segmentation Pavement Physical Characteristics Crash Prediction Models Run , Off , road Accidents Functional Forms Nonlinear Negative Binomial Regression