Abstract: This paper studies the prediction model based on integrated learning Xgboost to break the low accuracy limitations of traditional prediction methods in small sample data e-commerce products. It comprehensively considers the multi-dimensional indicators of e-commerce products, including：online search, online reviews, page access, inventory and order quantity, sentiment index etc., entropy method is used to fuse the same type of indicators. Logistic function and regular correction term are applied and a sales forecasting model based on integrated learning Xgboost is built combining the greedy algorithm. Carrying out model test for lenovo zuk z2 mobile phone of Jingdong Mall, and comparing the results with BP neural network prediction model, SVM support vector machine prediction model and BP-SVM combination forecast model, the result shows that the accuracy of the Xgboost prediction model with multidimensional index fusion is higher, and this study provides new method and idea for e-commerce products sales forecasts under small sample data.

Key words: sales forecast, e-commerce products, small sample, multiple indicators fusion, Xgboost

摘要： 为突破传统预测方法在小样本数据下电商产品销量预测中精度较低的局限，开展基于集成学习Xgboost的预测模型研究。综合考虑影响电商产品销量的多维指标，包括：在线搜索、在线评论、页面访问、库存与订购量、情绪指数等并利用熵值法融合同类指标。应用Logistic函数和正则修正项，结合贪心算法划分子树，构建基于集成学习Xgboost的电商产品销量预测模型。针对京东商城的联想zuk z2手机产品进行模型检验，并与BP神经网络、SVM支持向量机、BP-SVM组合预测三个模型进行对比，发现融合多维指标的Xgboost预测模型的精度显著提高，为小样本数据下电商产品销量预测提供方法和思路。

关键词: 销量预测, 电商产品, 小样本, 多维指标融合, Xgboost