Prediction of Used Car Price Based on LightGBM

Abstract

China’s used car trading market has shown a booming trend in recent years. Scientific evaluation of car prices allows buyers and sellers to trade used cars at reasonable and fair prices, which helps to reduce the transaction risks in the used car market and promotes the healthy development of the used car market. Although much work has been done on used car price prediction, challenges remain for accurate and robust prediction. This paper proposes a more cutting-edge machine learning algorithm based on LightGBM to predict used car prices authoritatively and innovatively using actual transaction records of a used car trading platform. Firstly, the original dataset is cleaned, and the features are filtered by analyzing the importance of each feature. The filtered features are then fed into LightGBM, and the parameters of LightGBM are tuned using grid search technology to improve prediction accuracy. Extensive experimental results have demonstrated that our proposed used car price forecasting algorithm has better prediction accuracy compared with classical linear regression, SVM, RandomForest, GBDT, XGboost, and other algorithms.