Research of the Correlation Between the Number of Training Data and Machine Learning Prediction Accuracy in Vehicle Crash Analysis

Abstract

In a vehicle crash analysis which has a strong non-linearity, the prediction accuracy of machine learning (ML) using FE simulations data is considered to depend on the number of training data. In this paper, the correlation between the number of training data and prediction accuracy of machine learning was researched using a large number of parametric FE simulation data in the small overlap frontal crash with fractures in vehicle structure.