基于分布式混合数据挖掘的电信客户流失分析  

Telecommunication Carriers Customer Churn Analysis Based on Distributed Hybrid Data Mining

作　　者：李爱群[1,2] 乔晗[1] 王汝传[1,2] 邓松[1] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210003 [2]南京邮电大学计算机研究所,江苏南京210003

出　　处：《计算机技术与发展》2010年第10期43-46,共4页Computer Technology and Development

基　　金：国家自然科学基金(60973139;60903181;60773041);江苏省自然科学基金(BK2008451);江苏省级现代服务业发展专项资金(2010002);江苏省高校自然科学基础研究项目(09KJB520009);国家和江苏省博士后基金(0801019C;20090451240;20090451241);江苏高校科技创新计划项目(CX09B-153Z;CX08B-086Z);江苏省六大高峰人才项目(2008118);江苏省计算机信息处理技术重点实验室基金(2010)

摘　　要：CORBA技术庞大而复杂,且技术和标准的更新相对较慢。电信运营企业应用系统是客户流失分析的主要数据来源,而传统的客户流失分析由于该系统数据的集中式存储继而采用集中式挖掘,对海量数据的挖掘效率低下。为进一步提高挖掘效率,提出网格下基于分布式混合数据挖掘的电信客户流失分析(Customer Churn Analysis upon Distributed HybridData Mining in Grid,CCA-DHDM),并借助GridSphere门户,在该平台上实现了BP神经网络算法和K-Means聚类算法。仿真实验表明,与单机环境相比,随着网格节点数增加,算法的平均耗时明显下降65%到75%,同时算法的效率得以较明显地提高。CORBA is a large and complex technology,and the updating of technique and standard is relatively slow.Telecom enterprise application systems are the main source of data for customer churn analysis,the traditional customers churn analysis uses a centralized mining due to centralized data storage,and the mining efficiency for mass data is low.Present CCA-DHDM（Customer Churn Analysis upon Distributed Hybrid Data Mining in Grid）,and achieve the BP neural network algorithm and K-Means clustering algorithm in this platform by means of GridSphere Portal.The simulation shows that,compared with stand-alone environment,the average time-consuming of algorithm is decreased obviously 65 percent to 75 percent with the grid nodes increasing,and the efficiency of the algorithm is improved clearly.

关 键 词：客户流失分析 网格计算 BP神经网络 K-MEANS聚类算法 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]