顧客關係管理系統(Customer Relationship Management System,CRM)在現在這個競爭激烈的商業環境中漸漸受到重視;以相關研究來說,取得新顧客的成本是維持舊有客戶費用的五倍,而能使舊有客戶的流失比率降低5%,獲利就可以增加60%以上。又以80/20法則,一間公司80%的利益來自20%的客戶,所以如何能確切瞭解客戶需求,從而發展出對客戶與企業有利的價值、策略與機制,是目前企業相當重視的一個議題。 本篇論文以一無線網路公司三個月來的交易與連線紀錄資料作分析,希望能以資料採礦中的叢集演算法找出各種不同客戶使用行為習性的群集,以求能提出有效的促銷行銷策略;另外,更希望進一步以決策樹演算法找出可能流失客戶群的特徵,以求擬定策略預防客戶的出走。 The customer relationship management system (the Customer Relationship Management System, CRM) gradually is subjected to the value in the business environment of this competition vehemence of now; Say with the related research, the cost that obtain the new customer is to maintain old customer''s expenses of 500%, and can make the old customer run off the ratio to reduce 5%, the profit can increase 60% above. Again with 80/20 rules, a benefit with 80% company comes from 20% customer, so how the ability is accurate to understand customer''s need, thus developing the value, strategy and mechanism that is beneficial to the customer and the enterprise, is a subject that the enterprise rather values currently. This thesis records the data to make the analysis with a wireless three months of net company-old the bargain and on-line, hoping to provide for to anticipate to mine for minerals the flock that the medium clustering algorithm finds out various different customer the usage behavior temperament, in order to can put forward the valid promotion marketing policy; Besides, even hope to be further to find out the characteristic that may run off the customers with the decision tree algorithm, prevent customer from fleeing in order to the draw-up strategy.