Short-term Popularity Forecasting Method of Retail Products Based on Combined LSTM-XGBoost Model

Accurately predicting the future market popularity of retail products is very important for retail enterprises to perceive market trends in time, thereby optimizing inventory management and improving marketing strategies. This paper proposes a retail product short-term popularity prediction method based on the combination of long short-term memory network (LSTM) and extreme value gradient boosting (XGBoost) algorithm. It comprehensively considers the characteristics of multiple dimensions such as product historical sales data, market share, track competitiveness and brand influence. By combining the time series analysis ability of LSTM neural network and the generalization ability of XGBoost algorithm, it deeply explores the historical pattern of product sales and identifies the future sales trend of products, thereby realizing accurate prediction of the short-term popularity of retail products. This study was tested on an actual cigarette ordering dataset of the Chinese tobacco industry. The results show that the proposed combined LSTM-XGBoost model has high accuracy and reliability in product popularity prediction, and can effectively support the market response and operational decision-making of retail enterprises.