Predicting casualty-accident count by highway design standards compliance

Compliance to standard has been the main doctrine in highway design, but its relationship with accident count has not been widely scrutinized. One of the key programs in road safety in Indonesia is road-worthiness test which assesses the compliance of a road to national design specifications and criteria. In light of current improvements in the crash data system in Indonesia, this study is carried out to develop a model to predict the accident count per type of crashes and to identify significant road features based on their compliance to a national standard. 272,200 km of arterial road in East Java North Corridor (EJNC) is selected as case study and 2012-2014 crash data is analyzed. Zero-Inflated Negative Binomial (ZINB) Model is preferred to develop crash prediction model with significant variables.This study has several findings. First, the constant values of the prediction model are relatively close to the average number of accident which implies that the mere compliance to current standard cannot warrant the safeness of Indoensian highways. Second, the number of median opening per unit length and disturbance level to pedestrian and road reserve area are the features that having positive relationships with total accident count. Meanwhile, the ROW disturbance, conformance of intersection and of road marking also show significant value but negative relationship with total accident count. Third, significant variables for each type of crash may have different sign. For example, in right angle crash, median width has positive relation with the number of accident, while in run off and rear end crash, median width compliance is shown to have negative relation.