|
Citation
|
Ejima S, Goto T, Zhang P, Cunningham K, Wang S. 27th International Technical Conference on the Enhanced Safety of Vehicles (ESV); April 3-6, 2023; Abstract #: 23-0034, pp. 9p. Washington, DC USA: US National Highway Traffic Safety Administration, 2023 open access. |
|
Abstract
|
27th International Technical Conference on the Enhanced Safety of Vehicles (ESV): Enhanced and Equitable Vehicle Safety for All: Toward the Next 50 Years
https://www-esv.nhtsa.dot.gov/Proceedings/27/27ESV-000034.pdf
Advances in automotive telemetry technology have the potential to predict occupant severity from vehicle conditions at the time of an accident, and appropriate triage, as well as transport to a trauma center, can greatly improve subsequent treatment. The National Automotive Sampling System Crashworthiness Data System (NASS-CDS: 1999-2015) was used to filter for new case selection criteria based on vehicle type and matched to Subaru vehicle categories. The authors have proposed four types of injury severity prediction algorithms that were matched with the categories of Subaru vehicles. Specifically, 1) ISP model that categorized the principal direction of force (PDOF) into four impact directions (front, left, rear, and right) , 2) ISP-R model that considers the effect of the right-front passenger in addition to the four impact directions, 3) ISP-f1R model that represents PDOF as a continuous function using periodic basis splines, called functional data analysis, and 4) ISP-f2R model in which the knot position was modified in 3). In this study, five-fold cross-validation was performed within the training data (NASS-CDS 1999-2015) to evaluate the performance of these four models. In addition, external validation was performed using the National Automotive Sampling System Crash Investigation Sampling System (NASS-CISS: 2017-2019). The results of the cross-validation showed that the area under the receiver operating characteristic curve (AUC) was used to evaluate the model performance, which was 0.854 for the ISP model and 0.862 for the ISP-R model, indicating that the ISP-R, which considered the influence of the right-front passenger, was more accurate. The AUC values were 0.847 for the ISP-f1R model and 0.856 for the ISP-f2R model using a continuous function for the direction of impact, indicating that the ISP-R model had the highest AUC value among the models. On the other hand, the validation results with NASS-CISS were 0.817 for the ISP model and 0.828 for the ISP-R model, and 0.831 for the ISP-f1R model and 0.828 for the ISP-f2R model, indicating that all models had AUC values above 0.8. The important factors for the occupant injury prediction algorithm were delta-V, belt use, age, and crash direction, and the presence of a right-front occupant was a significant injury risk modifier, especially in side impact crashes.
Language: en |