Enterprise Inventory Demand Forecasting Based on K-Means Clustering and a BP Neural Network

Abstract

Against the backdrop of rapid e-commerce growth and increasingly personalized consumer demand, fast-moving consumer goods (FMCG) enterprises face the dual challenges of excess inventory and stockouts, which impose higher requirements on the accuracy of inventory-demand forecasting. Intensifying market competition, frequent promotional campaigns, and rapidly changing consumer preferences have made the forecasting task substantially more complex. This study uses daily sales data for the core products of Company Z, an FMCG enterprise, from 2023 to 2025. A multidimensional indicator system is constructed from product-demand characteristics, and K-means clustering is applied to classify the products. BP neural-network models are then used to forecast daily inventory demand over the following four weeks according to the demand patterns of each product cluster. The results show that, compared with the exponential-smoothing method currently used by the company, the proposed hybrid model achieves a closer fit to actual sales and reduces the root mean square error by an average of 42.3%, thereby demonstrating stronger accuracy and adaptability. The combination of K-means clustering and a BP neural network can effectively capture the nonlinear and volatile characteristics of FMCG demand and provide more accurate support for inventory planning, with considerable practical value and potential for wider application.