PREDICTING THE PRICE OF AUCTION CARS WITH MACHINE LEARNING ALGORITHMS

Abstract

The problem of the automotive auction market to estimate the price of the cars accurately becomes critical as the number of features and interaction between these features grows and the conditions are also not standardized. In the present study, three machine learning algorithms—Linear Regression, Random Forest Regression, and an Extreme Gradient Boosting (XGBoost)—are compared using a unique dataset that was developed by integrating past data from car auctions to predict the prices of cars at auctions. Specific data features for the domain were also added, such as make, model, manufacturing year, engine, mileage, exterior color, chassis code, package trim and standardized auction condition grades (1.0 through 5.0). All missing value imputations, label encoding, Z-scores normalization, and more complex feature engineering methods, such as Vehicle Age, Mileage Intensity, Luxury Brand Mapping, and Make-Model Interaction terms have been performed prior to the processing phase. Performance of models was measured by Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and R-squared (R²) measures. Experimental results show that the accuracy of prediction of XGBoost is observed to be the highest with R² = 96.68%, MAE = 1,403, and RMSE = 1,975 which is higher than the accuracy of Random Forest (R² = 0.9527) and Linear Regression (R² = 0.8321). The results confirm the previous findings that ensemble-based gradient boosting methods improve considerably against linear models abilities when the price estimation is a dedicated task of an auction domain, particularly when feature engineering is employed to enhance the abilities