幾年來，資料挖掘(Data mining)技術受到許多企業的重視，更被廣泛地應用到顧客關係管理中，尤其是具智慧型的挖掘技術，例如類神經網路(Artificial neural networks, ANNs)中的非監督學習模式(unsupervised learning)，透過類神經網路的學習特性從資料中來挖掘出有用的資訊是它和傳統的演算法的不同；而在挖掘的過程中，由於種種可能的原因，都會有所機會發生資料遺漏或是不完整的情況，以往使用者只能捨棄該筆資料，此情形越多其挖掘出來的資訊偏差率就越大。因此，陸續有研究學者開始研究討論此相關領域的問題，最常見的方法為直接刪除、平均值或是眾數來取代等，但這種近乎直覺式的取代法對於最後的決策行為無法提供有意義的參考資訊。 本研究改以群組的「物以類聚」特性原理來思考，以此觀點來尋求獲得遺漏值較適推估值。而類神經網路法中自組織映射圖網路模式，正是一個常見且發展多年的神經網路模式，其特點則是對於輸入值的資料型態並無任何的限制，所以廣受各領域學者的歡迎，其成效也令人滿意；因此，本研究嘗試利用自組織映射圖網路來建立一個針對遺漏值問題的推估模式架構，透過自組織映射圖網路的群聚方法來找出遺漏值最適推估值，讓使用者可以在引用資料挖掘法時仍能保有最大的資訊量，期使挖掘出的結果更有意義。並且以資料庫（RFM資料庫資料）以及工業製程（半導體銅製程實驗資料）兩種類型的應用問題來展現本研究方法的可行性與合理性。 Recently, the technique of data mining had been applied into the issue of customer relationship management (CRM) by most enterprises. Among those techniques, the artificial neural networks (ANNs) had been mentioned as an intelligent approach. Especially, the unsupervised learning mode during ANNs can mine the available information by learning the clustering characteristic from data. However, the data may be lost or incomplete (i.e. the missing value) as for some particular reasons. To delete the missing value since making decision analysis is frequently employed to do it. Besides, the replaced estimate, e.g. the average value or the mode value, will be another approach to manipulate this problem. No matter what approach we take, the information we got will be limited for our decision analysis. In this study, we take the logistic concept of “clustering” to deal with such problem. The self-organize mapping (SOM) model of ANNs, which had been mentioned well for many applications, with the unsupervised learning will be taken to construct our solution herein. Finally, two illustrative examples including the missing value for database and for the experimental design are employed to demonstrate the rationality and availability for our proposed approach.
