A Deep Learning Approach for Used Car Price Prediction

Abstract

Buying a used car can be a challenging experience. Like many other consumer goods, used car prices have risen rapidly in recent years. In addition, gasoline prices and rising interest rates have made the experience of owning a car even more painful. In this research, we propose an intelligent framework for estimating the cost of used cars using artificial neural network algorithms. The model was developed using a training dataset of 140,000 used vehicles from 30 popular US car brands. The model's predictions are validated against a test data set of 35,000 used cars. Numerous features are examined for reliable and accurate predictions. Artificial neural networks are built using the Keras regression algorithm, and their performance is compared to basic models such as linear regression, decision tree algorithms, gradient boosting, and random forests. Categorical variables were processed using embedding techniques to improve predictive performance. The results are consistent with actual values and significantly improved over the baseline model. Experimental results showed that an ANN model with a mean absolute percentage error of 11percent and an R2 value of 0.96 outperforms the random forest model with a MAPE of 14 percent and an R2 value of 0.94.