Logo image
Back
Identifying Important Risk Factors Associated with Vehicle Injuries Using Driving Behavior Data and Predictive Analytics
Conference proceeding

Identifying Important Risk Factors Associated with Vehicle Injuries Using Driving Behavior Data and Predictive Analytics

Michal Monselise, Ou Stella Liang and Christopher C Yang
2019 IEEE International Conference on Healthcare Informatics (ICHI), pp 1-10
Jun 2019
DOI: https://doi.org/10.1109/ICHI.2019.8904860

Abstract

big data driving behavior driving disengagement injury prevention naturalistic driving predictive analytics secondary tasks vehicle injuries
Road injuries are rated among the top 10 causes of death by the World Health Organization, and the only one that is not a disease. The total economic cost of motor vehicle crashes in the United States was estimated to be 242 billion a year. This study examines multiple factors of accidents simultaneously with a goal of generating an interpretable model that can predict the occurrence of an accident given road conditions and driver behavior. The study compared 4 machine learning and deep learning modeling techniques on a dataset of 7707 trips collected by the Second Strategic Highway Research Program. A gradient boosted model was found to be most accurate and interpretable in accident prediction. This modeling technique also allows us to rank the feature importance of the factors in the model. The study finds that driver behavior, pre-incident maneuvers and secondary task duration are the most important variables in the predictive model. Using these conclusions will allow us to perform more work to infer these accident causes directly from vehicle sensor data in the future.

Metrics

4 Record Views
5 citations in Scopus

Details

Logo image