Modeling e-commerce retailer demand using Stacking Ensemble and K-means clustering patterns.

Abstract

Demand forecasting plays a crucial role in e-commerce. By accurately forecasting product demand, e-commerce companies can better manage inventory levels and decrease excess or out-of-stock situations, thereby reducing inventory costs and increasing customer satisfaction. To address the issue of the traditionally low accuracy in market demand forecasting by practitioners who relied on manual experience, this paper proposed a demand forecasting model for e-commerce retailers based on a Stacking ensemble model. The paper successfully modeled and analyzed past user behavior data from JD.com platform, and the model achieved automated forecasting of future demand based on historical data. Upon testing, the proposed Stacking ensemble model demonstrated RMSE (Root Mean Square Error) is 256.34 and 1-WMAPE is 0.916 close to 1, indicating that the model performed well and was capable of accurately predicting the demand for these products. Additionally, this paper also employed a K-means clustering model based on a cost function to classify products into five distinct categories according to their demand characteristic indicators and product attribute indicators.