Examination of Correctness and Wrongness Judgment Methods of Behavior Prediction in Driving Behavior Prediction Distribution by Bayesian Deep Learning

Abstract

In this study, we investigated a judgment method that classifies the prediction of driving behavior as correct or incorrect using a histogram of the prediction probability of driving behavior. XGBoost was used to judge the correctness of the prediction of driving behavior. The relative frequency of the histogram and the prediction class of driving behavior were used as the input features to XGBoost. The evaluation results of the judgment method for predicting driving behavior showed that the true negative rate was 63 %. This suggests that the proposed method is effective when incorrect predictions need to be excluded as much as possible.