A Study on Injury Prediction Method and Influential Factors in Side Impact Crashes Using Accident Data

Abstract

In order to reduce casualties in traffic accidents, more analyses are required from the engineering and medical science viewpoints when investigating traffic accidents. A key technology in this respect is injury prediction, which reduces the response time of the emergency rescue teams through automatic collision notification, and evaluates the injury mechanisms and risks. In this study, research questions are clarified by studying the factors that influence side impact crashes in traffic accidents in Japan, and determining significant factors that reduce accident casualties by constructing an injury prediction method. In this study, statistical accident analysis is applied to a combination of data obtained from the Institute for the Traffic Accident Research and Data Analysis (ITARDA) in-depth accident investigation database (Micro Data) and ITARDA Macro Data which is the official database for traffic accidents that occur throughout Japan. First, 73 nearside and 108 farside crashes are selected from Micro Data. Second, an object variable and 22 explanatory variables from the accident information are selected and categorized. The object variable is the maximum of AIS (MAIS). Third, an ordinal logistic regression analysis is performed to construct the injury prediction models using these variables. Finally, a comparison of the obtained results with Macro Data validates these models, and the outlier cases show extraordinary accidents. The results show that the significant factors are delta-V, damage grade, and striking vehicle weight for nearside crashes; and delta-V, vehicle category, lap zone-vertical, and seat belt use for farside crashes. Thus, these two types of crashes have different factors. The MAIS predictive accuracy rate is approximately 75, and the judgment rate of fatal and serious injuries (MAIS 3+) is approximately 90 %. The prediction models can correspond to Macro Data. The accident conditions arising from ±1 MAIS residual exhibit several characteristics. Eight cases with a large degree of error in nearside crashes involved five cases of SUV or truck side impact crashes with serious chest injuries. In farside crashes, 12 cases with a large degree of error included four unbelted drivers with serious head or neck injuries. These cases indicate that these extraordinary accidents have additional factors that influence injuries. At present, there is no published research on the side impact prediction method using Japanese Micro Data and Macro Data. In the United States, several injury prediction methodologies have been studied. However, only a few studies refer to accidents with residuals. In conclusion, this study shows the results of the proposed injury prediction methods and significant factors in side impact crashes by using data on traffic accidents in Japan. These prediction models can correspond to documented accident data. Furthermore, the residual analysis conducted indicates characteristic accident factors such as SUV side impacts for nearside crashes and unbelted drivers for farside crashes. Further investigations into these outlier accidents would prove beneficial in reducing traffic accident casualties in Japan.

F2012-F05-003