隨著網路通訊的蓬勃發展，便捷且收費低廉的網路電話成為用戶之間極受歡迎的溝通管道。然而根據統計此溝通平台却廣泛存在許多的詐騙或異常使用行為，帶給使用者嚴重的困擾與損失，對網路的安全管理及網路的品質也造成了很大的挑戰和負擔，如何有效偵測管制網路電話異常行為已成為電信管理重要研究課題。有鑑於此，本研究提出一個整合式網路電話異常偵測機制，期能利用此機制，找出可疑的異常行為模式，提供相關管理者進行管理監控。其主要架構分為三個階段進行偵測，首先利用模糊分群法將通訊記錄進行分類，然後利用案例式推理對分類特徵進行診斷和比對，過濾出可能之異常或詐騙行為，最後則利用規則分析技術實質來強化偵測異常行為能力。模擬研究發現，經由本研究機制，可有效偵測出網路電話之異常行為模式，過濾部份非監督式異常案例。研究結果，期能未來應用於網路電話異常偵測管理，並進一步減少詐騙通訊案例發生。

With the rapid development of network communication, convenient and low-cost Internet phone become a very popular channel of communication between users. However, according to the statistics of this communication platform has widespread fraud or abnormal behavior, give the user a serious distress and burdens and loss of network and caused a lot of challenges of security management and network quality, thus how effective detect abnormal behavior of control VoIP telecommunications management has become an important research topics. For this reason, this study proposes an integrated VoIP anomaly detection mechanism can take advantage of this mechanism to identify possible abnormal behavior patterns, and provide relevant managers to management control. Its main structure is divided into three stages of detection, the first use of telecommunications data classification, then use the case-based reasoning to diagnostic classification characteristics to filter out possible anomalies or fraud, and finally use the rule analysis techniques to cluster capacity of detection of abnormal behavior. In this empirical study, the mechanism can effectively detect the abnormal behavior of VoIP and unsupervised filter part of the abnormal cases. The findings can be used in future applications of VoIP anomaly detection, and further reduce fraud communication events.