A Hybrid Deep Learning Model for Car Sales Forecasting

Abstract

With the rapid development of the automotive industry and the continuous changes in market demand, accurate car sales forecasting has become an important tool for enterprises to optimize decision-making. However, due to limitations in data quality and the complexity of time series patterns, car sales forecasting still faces many challenges. To address these issues, this study proposes a hybrid deep learning model that integrates the Extreme Gradient Boosting (XGBoost), Long- and Short-term Time-series Networks (LSTNet), and Convolutional Neural Networks (CNN) based on car sales and production data from 2004 to 2023. This model utilizes XGBoost to extract key time series features, captures long-term dependencies through LSTNet, and efficiently fuses the predicted sequences generated by XGBoost and LSTNet with CNN. The experimental results show that the proposed model significantly outperforms traditional methods in prediction accuracy, providing more reliable decision support for enterprises. Model predictions show that by 2030, China's automobile sales will reach 26.77 million units, with BYD, Volkswagen, and Toyota expected to lead the market.