Advanced Predictive Analytics for Used Car Pricing Through Convolutional Neural Networks

Abstract

The automotive industry's explosive growth has led to a surge in the used automobile market, making it increasingly challenging for buyers and sellers to determine precise prices. This paper employs a Convolutional Neural Network (CNN) method to develop a reliable machine-learning model for predicting the price of used automobiles. A CNN-based approach is used to capture all the patterns in numeric or categorical data forms to determine car prices. The dataset applied in this work contains records of thousands of car descriptions and their images. On this dataset, different techniques of CNN model training, validation, and testing are done with much higher accuracy than the basic methods. The proposed system has better results than the conventional machine learning models by using the CNN model to employ feature extraction to obtain high accuracy in discovering relations between the features. The performance of the proposed CNN model is further compared with that of the Random Forest (RF) model. The experimental result of the CNN model shows R² of 0.85%, MSE of 3.65 %, RMSE of 1.91%, and MAE of 1.056 %. This can be widely applied to other areas where both quantitative data and images contribute to price determination, thus, the applicability of this model in predictive analytics is comprehensible.