本研究以資料探勘常見理論為基礎，建構出防堵垃圾郵件機制。使用PHP程式語言擷取電子郵件特徵，並透過機率類神經網路演算法、貝式分類器與C&RT (Classification and Regression Tree)對電子郵件分類，比較其分類模式之優劣。若考慮兩種可能狀況之下，發現設定平滑參數為0.01、0.1之機率類神經網路表現最好，其次貝式分類器與C&RT。也透過統計方法的變異數分析與Tukey真實顯著差異多重比較客觀分析其分類模式之優劣，發現與之前所做之結論一致。此外也使用風險分析，提供使用者在電子郵件分類不同的概念，評估分類模式是否符合使用者的需求。最後加入關鍵字搜尋，針對郵件主旨及寄件者名稱，建構黑白名單過濾，再配合機率類神經網路對電子郵件分類，看其評估準則是否提升。 The purpose of the study is based on the common theory of data mining that build up the mechanism of anti-spam. Using PHP program to pick the character of spam mail, it performs probability neural network (PNN), classification and regression tree (C&RT) and naïve bayes classifier to the E-mail classification, and compares three kinds of classified patterns. If considers under two kind of possibilities conditions, the probability neural network of smooth parameter 0.01, 0.1 is best, next C&RT and naïve bayes classifier. Using the statistical method of one way ANOVA and Tukey Multiple comparison test, 0bjectly it fits and unfits qualities of classified pattern that is consistent with the front conclusion. In addition, it uses cost of risk that provides the user in the email classification different concept and evaluates the three of classified patterns whether conforms to user''s demand. Finally, it joint the method of keyword search that aim at the field of subject and from to construct white-list and black-list, then to use PNN to E-mail classification whether increasing accuracy rate.
