Predicting Used-Vehicle Resale Value in Developing Markets: Application of Machine Learning Models to the Kazakhstan Car Market

Abstract

The burgeoning trade in used vehicles has necessitated further research into price prediction. In developing nations, the abundance of second-hand cars and limited supply of new ones has led to a preference for used vehicles. Consequently, the analysis of vendor data becomes imperative for gaining valuable insights. Sellers are increasingly seeking accurate price predictions to maximize their profits. The assessment of used car prices necessitates a thorough understanding of the features that influence value. Although the inclusion of multiple features can enhance prediction accuracy, the list of these features is non-exhaustive. This study seeks to examine the effectiveness of various regression techniques such as Linear, Decision Tree, SVM machines, Neural Network, and Bagged Trees, alongside machine learning algorithms, in predicting the selling price of used cars based on the associated features. Evaluation metrics will be utilized to identify the most proficient model by examining the performance and error rate of each model. The deep neural network model demonstrates exceptional performance, as indicated by its low RMSE and MSE values, suggesting high efficiency. Some models, including cubic SVM, fine Gaussian SVM, and wide neural network, exhibit a robust correlation (R) in accurately connecting input and output variables. Furthermore, narrow, medium, bilayered, and trilayered neural networks display commendable performance in recording variable correlations. After comparing various models, Bagged Trees were identified as the most cost-effective option per square meter, due to their advantageous pricing and performance.