Used Car Price Prediction Using K-Means Clustering and Comparison of Four Supervised Learning Regression Algorithms

Abstract

The Used car price prediction is a crucial challenge in the automotive industry as the need for an accurate and objective valuation system increases. This study aims to implement and compare the performance of four regression algorithms, namely Multiple Linear Regression (MLR), Polynomial Regression (PR) degree 2, K-Nearest Neighbors Regression (KNNR), and Random Forest Regressor (RFR) in predicting used car prices based on price segmentation using K-Means Clustering. The dataset used is the Used Cars Dataset from India containing 14,993 entries and 11 vehicle features. Segmentation is performed using the K-Means algorithm with the Elbow Method and Silhouette Score approaches, which show optimal results at k = 2 with a Silhouette Score value of 0.7170. The first cluster includes cars with a price range between 15,000 and 1,183,000 with an average of 514,411, while the second cluster includes prices between 1,185,000 and 2,271,000 with an average of 1,855,157. The results of this segmentation are then used as additional features in the regression model training process. Performance evaluation using MAE, RMSE, R², and MAPE shows that Random Forest Regressor provides the best results, with MAE = 92,439.34, RMSE = 146,838.26, R² = 0.9477, and MAPE = 18.25%, which indicates a high level of prediction. The integration of price segmentation techniques and non-linear regression models is proven to significantly improve prediction performance. The results of this study can be a basis for developing a more adaptive and precise data-based used vehicle price recommendation system.