The first and most important step in road safety management is the identification of roads with safety needs. One of the possible road screening techniques is based on the quality control method. Exposure data such as Annual Average Daily Traffic (AADT) and vehicle miles traveled (VMT) are usedto estimate the expected numbers of crashes, which are then compared to the actual numbers of crashes in order to identify locations with safety needs. This approach must be modified when applied to county and city roads because of the limited availability of exposure data. It is not expected in a foreseeable future that local transportation agencies establish traffic counting programs to cover all their roads. This paper proposes estimating the expected number of crashes on county roads based on surrogates of exposure data including: (1) segment connectivity with busier state-administered roads and (2) characteristics of the land development in the area. The land development data are included in the census database and they arewidely used in transportation planning. A method of estimating the expected crashes based on the classification tree is proposed. The method application to screen Indiana county roads is presented. The developed classification tree is compared to the benchmark model - Negative Binomial regression further improved by adding random effects. The proposed classification tree performs better than the benchmark model. The presented research demonstrates a promising method of screening local roads in a defendable manner.
Abstract