基于Stacking集成学习的商品销量预测研究  

Research on Commodity Sales Forecasting Based on Stacking Ensemble Learning

作　　者：沙通 代丽[1] SHA Tong;DAI Li(School of Economics and Management,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学经济管理学院,浙江杭州310018

出　　处：《物流工程与管理》2024年第4期116-119,60,共5页Logistics Engineering and Management

摘　　要：销量预测准确度的提升能够帮助电商企业更有效地规划库存,从而提高供应链管理的效率。文中利用Stacking集成学习方法融合多模型来预测销量,研究通过对基学习器进行选择,以RF、SVM、XGBoost、LSTM算法为第一层模型,线性回归为第二层模型,通过历史销售数据集进行验证。研究发现,相较于单一模型,Stacking集成学习方法的预测效果更优。此研究方法可以有效融合多种模型优势,得到更准确的销量预测结果,从而为电商企业制定库存和生产计划提供依据。Improving the accuracy of sales forecasting can assist e-commerce enterprises in more effectively planning inventory,thereby enhancing the efficiency of supply chain management.This article employs the Stacking ensemble learning method to integrate multiple models for sales forecasting.The research focuses on the selection of base learners,using RF,SVM,XGBoost,and LSTM algorithms as the first-layer models,with linear regression as the second-layer model.Through the validation with historical sales datasets,the study reveals that,compared to individual models,the Stacking ensemble learning method produces superior predictive performance.This research approach can effectively combine the strengths of various models,and yield more accurate sales forecasting so as to provide a basis for e-commerce enterprises to formulate inventory and production plans.

关 键 词：销量预测 Stacking集成学习 LSTM XGBoost 

分 类 号：F224.9[经济管理—国民经济] TP181[自动化与计算机技术—控制理论与控制工程]