顧客關係管理(Customer Relationship Management；CRM)近幾年在企業管理領域上造成相當大的話題，由於顧客消費意識的抬頭，使得企業與顧客之間的關係有所改變。以往「產品導向」之經營思維，逐漸不符合現今整體環境的需求，因而逐漸被更重視客戶感受之「顧客導向」的經營思維所取代。企業之行銷方式也由以往之「單一化、大眾化」，轉變為「個人化、客製化」之行銷方向。
基於如此的改變，企業需要透過分析以了解客戶的價值及需求。了解客戶的價值後，對於企業具高價值的客戶，其應該採取適當之保留策略以避免客戶流失；對於低價值的客戶，則須衡量所耗費之成本，避免浪費不必要之企業資源。
因此在本篇論文中，主要利用資料探勘中的叢集演算法對顧客做客戶區隔的動作。對於不同群集之使用者，分析其使用行為特徵，以利企業決策者制定適當之行銷策略以保留高忠誠度之客戶，並發掘潛在客戶群。
除此之外，針對具較高流失比例之叢集，本論文利用決策樹演算法分析其使用特徵。根據分析產生之結果，提供企業做為擬定防止顧客流失策略之參考

In recent years, Customer Relationship Management (CRM) has gained much attention on business administration. The relationship 
Between customers and corporations has changed because of the increasement of consumer awareness and knowledge. Customers now are facing huge amount of options from which to choose. Thus, more customers frequently change from one service provider to another in search of better service, which is called "churn".
In such competitive business environment, CRM plays an important role in enterprises. Many corporations have adopted Customer-oriented policy rather than product-oriented strategy nowadays. The marketing strategy also shifts from simplex and popular marketing to customize and personalize marketing.
In this paper, we applied customer segmentation by using clustering techniques in data mining. Customer segmentation involves identification of groups of customers with similar characteristics. According to the result, it could help enterprise to develop advisable business strategy and optimize the efficient use of resources throughout the organization. Besides, we also applied decision tree analysis to the clusters with high churn rates. Thus help enterprise preventing customer from leaving.

目錄 I
圖目錄 II
表目錄 V
第一章 緒論 - 1 -
1.1 前言 - 1 -
1.2 研究動機與目的 - 2 -
1.3 論文架構 - 4 -
第二章 相關背景知識 - 5 -
2.1 無線網路 - 5 -
2.2 資料探勘 (Data Mining) - 15 -
2.3 客戶關係管理(Customer Relationship Management) - 34 -
第三章 研究方法 - 45 -
3.1 研究架構 - 45 -
3.2 分析作業流程 - 50 -
第四章 實作分析結果 - 57 -
4.1 分析目標 - 57 -
4.2 基本統計分析 - 59 -
4.3 叢集分析結果 - 64 -
4.4 決策樹分析結果 - 83 -
4.5 流失分析結果之驗證 - 86 -
第五章 結論與未來研究方向 - 89 -
5.1 結論與未來方向 - 89 -
參考資料 - 90 -
附錄  英文論文 - 92 -

圖目錄
圖2.1 知識發掘之過程 16
圖2.2 資料探勘流程圖 17
圖2.3 絕對分割 24
圖2.4 叢集演算法基本流程圖 29
圖2.5 客戶生命週期 39
圖2.6 CRM基本運作架構 40
圖3.1 CRISP-DM 模型圖 46
圖3.2 分析作業流程圖 52
圖4.1 訓練與驗證資料集之選擇 57
圖4.2 各方案一般會員購買人數比例 59
圖4.3 各月份總上線人次統計 60
圖4.4 各月份月租用戶及上網卡用戶上線人次統計 60
圖4.5 各月份網路接收量及傳送量統計 61
圖4.6 各時段上線人次統計 62
圖4.7 各時段網路接收量統計 63
圖4.8 二元變易統計 67
圖4.9 續用客戶欄位分布瀏覽 68
圖4.10 流失客戶欄位分布瀏覽 68
圖4.11 目標叢集總覽 69
圖4.12 叢集[2]1欄位總覽 70
圖4.13 Frequency，當月使用次數 71
圖4.14 Hist_Avg_Frequency，過去平均使用次數 71
圖4.15 Duration_Hour，使用時數 71
圖4.16 Hist_Avg_Duration_Hour，過去平均使用時數 72
圖4.17 Workday_Time1_Ratio 72
圖4.18 Workday_Time2_Ratio 72
圖4.19 Workday_Time3_Ratio 72
圖4.20 Workday_Time4_Ratio 72
圖4.21 上線最大間隔天數 73
圖4.22 上網地點數/上網次數比率 74
圖4.23 戶外AP/室內AP上網比率 74
圖4.24 時段2(上班時段)使用AP數 75
圖4.25 叢集[4]3，4.87% 77
圖4.26 叢集[6]5，4.44% 79
圖4.27 叢集[7]6，2.69% 81
圖4.28 決策樹分析-利用Gap_Final_To_End 84
圖4.29 決策樹分析-利用Frequency 84
圖4.30 決策樹分析-利用Drain_Flag 85
圖4.31 流失叢集特徵規則 86

表目錄
表4.1 欄位時段切割說明 64
表4.2 叢集演算法作用欄位說明 65
表4.3 叢集演算法補充欄位說明 66
表4.4 叢集4[3] 78
表4.5 叢集6[5] 80
表4.6 叢集7[6] 82
表4.7 決策樹分析法使用欄位說明 83
表4.8 流失特徵規則說明 87
表4.9 流失特徵規則之驗證 87

[1] Anand Balachandran, Geoffrey M. Voelker and Paramvir Bahl P. Venkat Rangan, "Characterizing User Behavior and Network Performance in a Public Wireless LAN," 
[2] C. Archaux, A. Martin and A. Khenchaf, "An SVM based churn detector in prepaid mobile telephony," in 2004, pp. 459-460. 
[3] Corinne Baragoin, Christian M. Andersen, Stephan Bayerl, Graham Bent, Jieun Lee and Christoph Schommer. (2001, September). Mining your own business in telecoms using DB2 intelligent miner for data. International Business Machines Corporation, 
[4] L. Guo, M. Zhang, L. Sun and Z. Wang, "Churn analysis model of securities business based on the decision tree," in 2006, pp. 6048-6051. 
[5] Kok Wai Wong, Chun Che Fung, T. Gedeon and D. Chai, "Intelligent data mining and personalisation for customer relationship management," in 2004, pp. 1796-1801 Vol. 3. 
[6] Luc A. Andriantiatsaholiniaina and Ljiljana Trajkovic, "Analysis of User Behavior from Billing Records of a CDPD Wireless Network," 
[7] Y. Min, S. Bin and Li Ming-fang, "A kind of classification and regression tree algorithm for unusual customers recognition in telecom trade," in 2004, pp. 561-564. 
[8] Pete Chapman (NCR), Julian Clinton (SPSS), Randy Kerber (NCR), Thomas Khabaza (SPSS), Thomas Reinartz (DaimlerChrysler), Colin Shearer (SPSS) and Rudiger Wirth (DaimlerChrysler). (2000, CRISP-DM 1.0 step-by-step data mining guide. 
[9] Tutorial-Reports.com. Wireless LAN (wifi) tutorial. 
[10] Yi-Ming Yang, H. Wang, L. Li, Tian-Yi Li, Wen-Min Li, Q. Yang, W. Lv and P. Huang, "Multi-dimensional model-based clustering for user-behavior mining in telecommunications industry," in 2004, pp. 1650-1655 vol.3. 
[11] W. Yu, D. N. Jutla and S. C. Sivakumar, "A churn-strategy alignment model for managers in mobile telecom," in 2005, pp. 48-53. 
[12] Y. Zhang, J. Qi, H. Shu and Y. Li, "Case study on CRM: Detecting likely churners with limited information of fixed-line subscriber," in 2006, pp. 1495-1500. 
[13] 丁一賢 and 陳牧言, 資料探勘. 滄海書局, 2005, 
[14] 張志崧, "電信產業之盜用偵測," June ,2004. 2004. 
[15] 洪維國 and 王小璠. CRM 運作架構及服務策略發展之研究. 
[16] 甘傑英. WiMAX標準與認證簡介 an introduction to WiMAX standard and certification. 系統晶片科技中心技術期刊 第4期 
[17] 謝幸宏, "無線網路客戶行為分析 Analysis of wireless customers' behavior," June ,2006. 2006. 
[18] R. 陳榮昌 Chen and Y. 林育臣 Lin. 群聚演算法之比較及群聚參數的分析與探討. Presented at 第三屆網際網路應用與發展研討會.