Construction and Application Research of Beer Category Sales Forecasting Model Based on Big Data Analysis for Supermarket X

Abstract

This study constructs a sales forecasting framework incorporating multi-source data based on beer category sales data from Supermarket X. The research collected beer sales data from 2020-2023, integrating temperature data, holiday information, and promotional activity data through an improved LSTM-based deep learning model. The study innovatively introduced Attention Mechanisms into the prediction model and proposed a dynamic weight allocation method for seasonal features, effectively enhancing the integration of heterogeneous data. Experimental results demonstrate excellent prediction performance across different time scales, achieving accuracy rates of 95%, 97%, and 98% for daily, weekly, and monthly forecasts respectively. Through practical validation in five demonstration stores, the model application improved inventory turnover by 25%, reduced stockout rates by 35%, and lowered operating costs by 15%. The research findings provide effective decision support tools for retail enterprises' refined operations management while suggesting future research directions regarding limitations in seasonal product forecasting and extreme weather response, offering practical reference for regional retail industry's intelligent transformation.