融入个体活跃度的电子商务客户流失预测模型  

E-commerce customer churn prediction model combined with individual activity

作　　者：琚春华[1,2] 卢琦蓓[3,4] 郭飞鹏[5] 

机构地区：[1]浙江工商大学计算机与信息工程学院,杭州310018 [2]浙江工商大学现代商贸研究中心,杭州310018 [3]浙江工商大学工商管理学院,杭州310018 [4]台州职业技术学院工商管理系,台州318000 [5]浙江经贸职业技术学院信息技术系,杭州310018

出　　处：《系统工程理论与实践》2013年第1期141-150,共10页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(71071141);国家教育部博士点基金(20103326110001);浙江省自然科学基金重点项目(Z1091224);浙江省教育厅科研项目(Y201225624)

摘　　要：为提高个体层次上客户流失预测的精确度,建立了融入个体活跃度的电子商务客户流失预测模型H-ULSSVM.该模型首先利用融入地域因素的启发式算法计算出最优阈值,并求出个体的活跃度,识别出正判客户和错判客户;在此基础上,考虑电子商务客户流失预测影响因素众多,提出了一种粗糙等价类属性约简方法提取出重要的客户流失预测指标,然后将降维后的正判客户样本送到非平衡最小二乘支持向量机进行学习和训练,进而利用得到的分类器对错判客户样本的客户流失状态进行判别.在某B2C电子商务平台客户样本的实证研究表明,该模型与其他方法相比,具有更好的效率和精确度.In order to improve the accuracy of customer churn prediction at individual level, an E- commerce customer churn prediction model combined with individual activity called H-ULSSVM was established. Firstly, it used heuristic algorithm which integrated into geographic factors to calculate the optimal threshold and obtain the degree of individual activity, identify the correctly identified customers and incorrectly identified customers. On this basis, considering a large number of impact factors exist in E-commerce customer churn prediction, a rough equivalence class reduction method was proposed to extract important index. The correctly identified customers were sent to learn and train in unbalanced least squares support vector machine, and then used the classifier to judge the status of the incorrectly identified customers. The empirical study on B2C E-commerce platform shows that this model has better efficiency and accuracy than others.

关 键 词：客户流失预测 启发式算法 非平衡最小二乘支持向量机 粗糙集 

分 类 号：F713.36[经济管理—产业经济] TP18[自动化与计算机技术—控制理论与控制工程]