CSD-LSSVR-Based Inventory Demand Forecasting for Warehouse-Distribution Integrated SMEs

Abstract

The problem of inventory demand forecasting is an urgent issue in the development of warehouse-distribution integrated small and medium-sized enterprises (SMEs), which is of great importance to meet the sales demand of customers and significantly reduce distribution costs. The study describes the inventory demand problem of small and medium-sized enterprises. Based on the analysis of compressive sensing denoising methods and manual prediction methods, a prediction model is constructed using LSSVR and CSD algorithms. The study conducts an experiment using real order demand data of seafood customers from a small and medium-sized enterprise integrating warehouse and distribution in Sichuan Province from April 3, 2019 to September 9, 2023, with a total of 775 records. The training and testing sets are divided in a 4:1 ratio. Data preprocessing includes filling missing values using linear interpolation, detecting and correcting outliers using Z-score method, and normalizing the data to the [-1,1] interval. The experimental results show that on the test set, the relative error (RE) of the CSD-LSSVR model is 0.0701, the mean absolute error (MAE) is 58.258, the mean square error (MSE) is 70.12, and the directional statistic (DS) is 0.688; The RE of the traditional SVR model is 0.1214, MAE is 106.25, MSE is 112.25, and DS is 0.435. This indicates that the CSD-LSSVR model significantly improves prediction accuracy and stability. The above results indicate that the CSD-SVR prediction model performs better in inventory demand forecasting. This model can be applied to predict inventory demand for small and medium-sized enterprises, providing more possibilities for the efficient development of e-commerce enterprises.