Abstract
To predict vehicle occupant injuries based on accident information obtained from a video-recorded drive recorder, injury prediction models were constructed by machine learning using information currently available from drive recorders and information that is expected to be acquired hereafter. Light Gradient Boosting Machine and Bayesian Networks were used for the machine learning models. The results showed that it was difficult to predict injuries with only the information currently available from a video-recorded drive recorder, while the prediction model with additional information expected to be available hereafter improved the prediction accuracy.
