Multi-Source Data Integration and Sales Forecasting for Chain Retail Enterprises Using ETL Techniques

Abstract

Chain retail enterprises accumulate a large amount of business data from sales, stores, promotions, customer flow, and time environment in their daily operations. However, differences in structure, inconsistent field definitions, and complex information relationships between different data sources affect the stability and accuracy of store sales forecasting. To address this issue, this paper constructs an ETL data fusion and store sales forecasting method for chain retail enterprises. This method first extracts, cleans, transforms, and loads sales records and store attribute information from the Rossmann Store Sales dataset, performing preprocessing operations such as missing value handling, outlier correction, categorical feature encoding, and continuous variable standardization. Based on this, static store attributes, calendar features, promotion status, customer flow information, and recent sales signals are integrated into a structured feature representation, and the time dependency in store sales changes is preserved through historical window construction. Subsequently, a sales forecasting model is built based on the fused daily store data to estimate future sales revenue. Comparative experimental results show that the proposed method performs well in terms of RMSE, MAE, MAPE, and R2, effectively reducing prediction errors and enhancing model fitting ability. Research shows that ETL data fusion can provide a more complete and reliable data foundation for retail sales forecasting, and the constructed methods can provide data support for chain retail enterprises' inventory management, replenishment planning, promotion arrangements, and operational decisions.