隨著資料的快速成長與大量累積，資料探勘已被廣泛應用於許多領域，例如：決策支援、詐欺偵測、市場分析、財務預測等等。針對各種不同資料特性與研究議題，已有許多方法與技術被提出，用以從大量資料中歸納出有用的資訊，屬性導向歸納法是其中一項重要技術。然而，現有的屬性導向歸納法存在著二個問題：第一，其只依據二個關鍵門檻值進行歸納，所提供的廣義知識只是資料庫的一個知識片段，若想獲得完整的歸納知識，必須重覆進行多次歸納；第二，現有方法僅關注正向資料，缺乏對負向資料的處理。針對此二項不足，本研究提出二種新的歸納方法，得以一次歸納並產生所有有趣的多階層正向與負向廣義知識。此外，真實世界有著各種不同的知識種類，除了上述正向與負向知識之外，資料庫中亦存在著具有異常誤差的稀少性資料，傳統資料探勘方法僅能偵測異常物件，無法解釋物件中真正發生異常的屬性。因此，本研究提出第三種方法，能從資料庫中挖掘出真正造成物件異常的最小屬性組合，稱之為可疑樣式。經由真實資料集實際測試與評量，證明本研究所提出的方法具可行性並能有效找出有用知識。 Data mining has attracted a great deal of attention in the information industry and in society due to its wide applicability in many areas. Many approaches have been proposed to generalize valuable information patterns and attribute-oriented induction (AOI) is one of the most important methods. However, existing AOI approaches encounter two problems. First, the AOI only provides a snapshot of the generalized knowledge, not a global picture. Second, it only mines knowledge from positive facts in databases. In this study, we proposed two novel methods to generate all interesting multiple-level positive and negative generalized knowledge at one time. Moreover, knowledge types are various in real world. In addition to the positive and negative knowledge, a dataset may include very rare, suspicious values, or the abnormal deviations. Existing researches focused only on the identification of outliers which possess the same dimensional space, what are the explicit anomalous knowledge hidden in the mined outliers is rarely addressed. This study proposed third approach to discover such suspicious knowledge. Both proposed methods have been verified for efficiency and effectiveness by using real datasets.
