Abstract
Rear-end collision is one of considerable accident types in Japan because many occupants are injured in this type of crash; however, most of their injuries are slightly injured. It is difficult to analyze regression models of rear-end collision using Japanese Micro Data with sampling limitations and few serious injuries. This study clarifies influential factors and injury prediction model in rear-end collisions using Japanese Macro Data, which is the police reported database for all traffic accidents that occur throughout Japan. The injury prediction method and influential factors are proposed using ordinal logistic regression model. The significant factors are delta-V, vehicle damage grade, seat belt use, occupant age, and vehicle category. The prediction model using these factors can correspond to the rate of fatalities and serious injuries in Macro Data; furthermore, it can show the good correlation to Micro Data. This model indicates the serious injury risk for belted occupants with delta-V of 60km/h or less, which is the majority of conditions in rear-end collisions, is 20% or less. The conditions with the risk of 50% or more are quite limited. This result shows the judgment of serious injuries using the injury prediction model needs many overtriage cases. Therefore, the balance between sensitivity and overtriage for the injury prediction model is important issue in rear-end collisions.
