一种基于属性约简和SOM的客户细分方法  

A Customer Classification Method Based on Attribute Reduction and SOM Neural Network

作　　者：郑永前[1] 于胜男[1] 王永生[1] 

机构地区：[1]同济大学机械工程学院,上海200092

出　　处：《工业工程》2011年第2期31-36,共6页Industrial Engineering Journal

基　　金：上海市自然科学基金资助项目(10ZR1431700);"863"高技术研究发展计划资助项目(2008AA04Z105)

摘　　要：客户细分是保险行业进行差异化营销的基础。由于知识冗余的存在,采用传统的聚类方法进行客户细分存在细分质量低的问题。为有效进行客户细分,提出基于属性约简和SOM的聚类模型。应用属性约简规则处理数据可有效识别冗余知识,找出关键属性;将关键属性作为SOM神经模型的输入,提高客户细分质量。以H保险公司作为实例,使用该模型进行客户细分,通过聚类结果比较,证明方法有效。In the insurance industry,customer classification is the basis for differentiated marketing.Due to the knowledge redundancy,the normally adopted traditional clustering method for customer classification suffers from low clustering quality.In order to achieve an effective classification,this paper proposes a clustering model based on attribute reduction and self-organizing feature map（SOM） neural network.By using attribute reduction rule,it can distinguish the redundant knowledge and effectively discover key attributes.Key attributes are used as inputs for learning by SOM neural network so as to obtain better clustering quality.The effectiveness of proposed method is verified through a case study from the H Insurance company.

关 键 词：属性简约 聚类分析 自组织映射神经网络 客户细分 

分 类 号：F274[经济管理—企业管理]