Prediction of Automobile Aerodynamic Drag Coefficient for SUV Cars Based on a Novel XGBoost Model

Abstract

A novel XGBoost model (GA_XGBoost) was established and studied in this paper in order to predict aerodynamic drag coefficient for SUV car accurately. After normalizing the input data, some important wind resistant areas of automobile are obtained by accompanying solution analysis which are helpful to filter out more impacts on aerodynamic drag predicting in the coming mutual information method during model establishment. The genetic algorithm is used to find the optimal hyperparameter value of XGBoost. The results show that compared with the original XGBoost model, the GA_XGBoost is capable to contribute 20% and 16% reductions on mean absolute error and mean square error individually. Moreover, the GA_XGBoost model also improves predicting accuracy on automobile aerodynamic drag coefficient than other models. The knowledge gained herein not only builds a novel model with satisfactory performance on predicting automobile aerodynamic drag coefficient for the SUV model, but also provides theoretical fundament on establishing predicating models.