Crash severity can be defined as the potential of a crash to cause an injury or fatality. In the National Automotive Sampling System – Crashworthiness Data System (NASS-CDS), the crash severity of a rollover is assessed by estimating the magnitude of maximum intrusion and crush in the damaged vehicle. Several studies have shown that the number of quarter turns and roof intrusion are significant factors influencing the injury outcome. These studies mainly investigate the relationship between injury severity and vehicle-, crash-, or occupant-related variables. The purpose of this study is to develop a model that uses both vehicleand crash-related parameters to estimate the rollover crash severity based on injury outcome.

In this study, the data mining technique called discriminant analysis is used to build a predictive model. Of the several rollover-related variables considered as candidate predictors, the maximum intrusion, number of quarter turns, and estimated distance from trip point to final rest position show significant correlations with the maximum abbreviated injury scale (MAIS) and hence are selected as predictors for the model.

Since one of the predictors, the estimated distance from trip point to final rest position, was introduced in the NASS-CDS data in 2006, this study is based upon two years (2006 and 2007) of data. To eliminate the confounding effect of external sources of injury, only non-ejected occupants are considered. The data is also screened to include only the maximum intrusion in the vehicle and the occupant with maximum MAIS in the vehicle.

The discriminant function is used in building the model. Given the specific values of the predictors for a rollover case, the final model predicts the injury outcome in rollovers as minor, moderate, and severe with sufficient accuracy. The model can be used to extract comparable rollover cases to understand injury mechanisms that can be used to develop vehicle crashworthiness countermeasures.