A Sales Prediction Method Based on XGBoost Algorithm Model

Abstract

Reasonable and accurate sales forecasting is an important issue for large chain stores. Forecasting short- and long-term product sales helps companies develop marketing strategies and inventory turnover plans. In today's ever-changing business environment, the application of artificial intelligence technology allows for more efficient processing of large amounts of data while taking into account many external factors such as the climate, consumer patterns, and financial situation. An XGBoost linear regression model for the Kaggle competition was trained using the dataset of Ecuadorian Favorita chain stores that was made available. The suggested prediction model seeks to address the seasonality and data scarcity issues. In the context of machine learning, producing several samples for both training and testing aids in our ability to assess the model's efficacy. The most popular technique for detecting overfitting and underfitting issues is to create various samples of data for training and testing models. The experimental findings demonstrate that the XGBoost linear regression model can reasonably provide scientifically based predictions for chain store sales and has a high prediction accuracy.