Sales Forecasting for Retail Business using XGBoost Algorithm

Abstract

The retail industry is continuously evolving with the expansion of sales channels and the diversification of product assortments. However, current forecasting methods, relying on simplistic statistical models, frequently encounter difficulties in adjusting to the dynamic retail environment. This limitation leads to challenges in accurately predicting sales volumes and frequencies. Consequently, there is a critical need to improve the accuracy and frequency of sales predictions to enable timely decision-making for business strategies. Through a comprehensive analysis of datasets spanning from 2019 to 2023, this study illustrates the substantial advantages of integrating eXtreme Gradient Boosting (XGBoost) to gain deeper insights into sales patterns. Results demonstrate a significant enhancement in prediction accuracy, with an average reduction of 29.23% in Mean Absolute Error (MAE) and 34.54% in Root Mean Squared Error (RMSE) compared to conventional methods. Furthermore, the adoption of XGBoost facilitates the transition from monthly to daily forecasting, thereby optimizing the efficiency of the prediction process. Retailers can optimize inventory management, devise effective marketing strategies, and ultimately maximize revenue. The findings emphasize the importance of embracing innovative approaches to address the challenges of a rapidly evolving retail landscape and drive sustainable business growth.