USFF: A Unified Sales Forecasting Framework with Fourier-Enhanced Decomposed Net

With the widespread adoption of big data science, the retail industry has seen an increasing demand for data-driven decision-making. As a critical task within the sector, product sales forecasting faces significant challenges. Traditional statistical methods are limited by their reliance on assumptions of data stationarity, while DL and ML techniques enhance accuracy but struggle to fully address complex features such as temporal dependencies, nonlinear trends, and external influences. No single method proves universally effective across all scenarios. To tackle these challenges, this paper introduces a Unified Sales Forecasting Framework (USFF) for vending machine sales prediction, which integrates statistical methods, machine learning, and deep learning techniques. By classifying data based on statistical features and employing multi-model fusion strategies, the USFF framework effectively captures the complex nature of sales data, significantly improving prediction accuracy by leveraging the strengths of various models. Furthermore, the paper presents a novel deep learning forecasting model, FEDNet (Fourier-Enhanced Decomposed Net), designed to address the dynamic dependencies and seasonal patterns inherent in long time series. FEDNet offers substantial advantages in capturing long-term dependencies and dynamic changes in time series data through the integration of reversible instance normalization, seasonal-trend decomposition, and Fourier decomposition.Experiments conducted on over 30 million real-world data points, along with five public datasets, demonstrate that the proposed framework and deep learning model outperform baseline methods across multiple evaluation metrics, confirming their effectiveness and reliability. The proposed framework and algorithm have been successfully implemented in a retail company in China, providing valuable business decision-making support for operational strategies.