VMD-SVM-based Retail Product Demand Forecasting Model

Abstract

This study addresses the issues of noise and volatility in retail inventory forecasting by proposing a hybrid model based on Variational Mode Decomposition (VMD) and Support Vector Machine (SVM). Existing methods often struggle to handle noise in time-series data, which affects prediction accuracy. To overcome this, we decompose the raw time-series data into multiple modes using VMD, effectively removing noise and capturing the primary trends. The processed data is then modeled and predicted using SVM. The experiments first compare various machine learning models on the raw data, followed by VMD preprocessing to assess its impact on model performance. The results show that VMD significantly improves prediction accuracy, with SVM's R² increasing from 0.4039 to 0.9214. This research provides an effective solution for retail inventory forecasting, demonstrating the advantages of VMD in handling complex time-series data.