摘要
采用某网上商场的2525名客户样本,构建了基于SMC和最小二乘支持向量机(LSSVM)的电子商务客户流失三阶段预测模型。首先应用SMC模型计算出客户活跃度,以0.5为阈值判断出客户流失状态,识别出正判客户和错判客户;其次将训练样本送入LSSVM进行训练和学习,进而对测试样本的客户流失状态进行判别,然后将误判客户样本输入最近邻分类器进行再判断。结果表明,与SMC模型、BP神经网络模型、LSSVM模型相比,三阶段模型对测试样本预测精度更高,是一种更有效和实用的分类方法,可为电子商务企业客户关系管理提供一个新的方法。
Taking 2525 customers in an e-shop as samples,this paper proposes a three-step integrated model of SMC and least squares support vector machines（LSSVM） for E-business customer churn prediction.Firstly,customers＇ active probabilities are obtained by using SMC model to identify customer churn status with the threshold of 0.5.The training and testing samples are formed by the correctly identified customers and incorrectly identified customers respectively.Then LSSVM trained with training samples is used to identify customer churn status of testing samples.Finally,the incorrect customers of testing samples are re-identified with a nearest neighbor classifier.Empirical results show that,compared with SMC,BP neural network and LSSVM models,three-step integration model is an efficient and practical tool for E-business customer churn prediction of testing samples,and supplies E-business enterprises a new method in customer relationship management.
出处
《中国软科学》
CSSCI
北大核心
2010年第6期186-192,共7页
China Soft Science
基金
国家自然科学基金项目(70471074)
广东省自然科学基金项目(9452902001004060)
国家博士后科学基金(20100470008)
关键词
电子商务客户流失预测
三阶段模型
SMC
最小二乘支持向量机
最近邻法
E-business customer churn prediction
three-step integration model
SMC
least squares support vector machines
nearest neighbor algorithm
