Automobile sales forecasting based on Special Zero-inflated Data

Car sales projections are important for manufacturers in terms of production adjustments and strategic planning. Traditionally, such predictions tend to focus only on brand performance across a wide range of markets. The LSTM model showed good prediction results. Due to the increasing importance of Market Segmentation, the demand for some car sales forecasts (PAS) has increased, and PAS data contains a large number of zero values and has significant cyclicality. This creates problems with traditional statistical methods and becomes difficult to capture the nuances of the data. To solve this problem, a method is proposed that combines short-and long-term memory (LSTM) networks and zero-expansion Poisson models. The latter is suitable for processing data with superfluous zero values and complex patterns. The mixed-loss zero-expansion LSTM system (Zim-LSTM) has also been implemented to improve robust and long-term forecasting capabilities for future pa. Zim-LSTM has advantages in modeling common zero values and stores historical information to facilitate the short-sighted task of predicting time series. Using a method of verifying sales data of real cars and comparing them with existing reference models, the results show that the accuracy and reliability of Zim-LSTM have improved, making it a promising solution for predicting PAS.