Analyzing crash injury severity for a mountainous freeway incorporating real-time traffic and weather data

This study focuses on developing crash injury severity analysis models for a mountainous freeway section. In addition to the data obtained from crash reports, real-time traffic and weather data were utilized. The introduction of real-time data would benefit model applications on crash injury severity prediction. Crash injury severity was classified as a binary outcome (severe and non-severe crashes) and random forest model was firstly estimated to select the most important explanatory variables associated with severe crash occurrence. Four most critical variables (snow season indicator, steep grade indicator, speed standard deviation, and temperature) were chosen by the random forest model as inputs for the modeling analyses. For the purpose of identifying actual relationships between severe crash occurrence and the chosen explanatory variables and enhancing model goodness-of-fit, a total of three models were developed to analyze crash injury severity: (I) fixed parameter logit model; (2) support vector machine (SVM) model with radial-basis kernel function to detect non-linearity; and (3) random parameter logit model with unrestricted variance-covariance matrix to account for individual heterogeneity and also to investigate potential correlations between the explanatory variables. The three models were compared based on the areas under the ROC curve (AUC) values and it was demonstrated that SVM model and random parameter model provide superior model fits than the fixed parameter logit model. Findings of this study demonstrate that real-time traffic and weather variables have substantial influences on crash injury severity, which could be utilized to predict crash injury severity. Moreover, it is important to consider possible non-linearity and individual heterogeneity when analyzing crash injury severity. In addition, potential applications of the modeling results, limitations of this study have been discussed. (C) 2013 Elsevier Ltd. All rights reserved.