基于改进XGBoost的电商客户流失预测

E-commerce Customer Churn Prediction Based on Improved XGBoost

基于客户价值模型、聚类算法与机器学习算法,提出一套面向电子商务大数据领域的客户流失预测方法并对其进行验证。采用随机森林方法对高维数据进行降维并选择特征变量,利用RFM价值模型,详细地对客户进行划分。由于电商企业对真阳性错误更为敏感,所以对XGBoost算法引入惩罚因子,并结合特征变量预测客户流失,以提高预测准确性。根据对国内某电子商务平台客户数据的预测结果表明,经过预先客户细分处理的预测结果效果明显更好,同时改进后的XGBoost算法较改进前的预测准确率、精确率、召回率分别提升了2.8%、3.8%、2%,即所提出的预测方法是有效的。

Based on customer value segmentation model,clustering algorithm and machine learning algorithm,a customer loss prediction method for e-commerce big data is proposed and verified.Random forest method is adopted to reduce the dimension of high-dimensional data and select characteristic variables.Customers are segmented based on RFM value model.Because e-commerce companies are more sensitive to true positive errors,a penalty factor is introduced into the XGBoost algorithm,and combined with feature variables to predict easy customer churn to improve the accuracy of prediction.According to the prediction results of customer data of a domestic e-commerce platform,the prediction result which after pre-segmentation processing are significantly better,and the prediction accuracy,precision and recall rate of the improved algorithm are improved by 2.8%,3.8%and 2%respectively,that is to say,the prediction method is effective.