Residual Value Prediction of Pure Electric Vehicles via Stacking Model Fusion

Accurate residual value prediction of pure electric vehicles (EVs) is crucial for the development of the used vehicle market and risk management in the automotive industry. This paper proposes a stacking model fusion framework that integrates XGBoost, AdaBoost, and CatBoost algorithms to improve prediction accuracy. A dataset comprising 5,882 EV transaction records from 2024 was collected and preprocessed through feature engineering, standardization, and correlation analysis. Experimental results show that CatBoost performs best among single models (RMSE <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=1.91$</tex>), the proposed stacking ensemble further reduces the RMSE to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1. 7 5}$</tex>, outperforming all base learners. The findings highlight the effectiveness of model fusion in capturing complex feature interactions and enhancing generalization capability for residual value estimation.