Used-Car Price Prediction Based on Machine Learning

Abstract

Used-car valuation is a typical nonlinear regression problem in which price is jointly affected by vehicle age, mileage, reference price, brand, model style, and market location. This paper proposes a machine-learning framework that covers data preprocessing, feature engineering, model construction, and comparative evaluation. Using 32,455 cleaned used-car records, Ridge regression, random forest regression, and gradient boosting regression are trained under an identical experimental protocol. Random forest achieves the best test-set performance, with an MAE of 2.029 in ten thousand CNY, an RMSE of 6.694 in ten thousand CNY, an $R^{2}$ of 0.962, and a MAPE of 14.79%. The visualizations show close agreement between predicted and observed prices for most samples, while larger deviations are concentrated in the high-price range. Feature-importance analysis identifies original price, production year, vehicle age, and mileage as the principal predictors. The study provides a structured and reproducible reference for used-car price prediction.