一個由建構的乏晰決策樹產生乏晰分類規則的新方法

Abstract

ID3學習演算法是由Quinlan在 1983年所提出來用以解決分類問題的方 法,它是屬於一種批次學習的策略. 幾種結合了乏晰集合理論的乏晰ID3學 習演算法亦被證明可以獲致較佳的分類效果.這是因為它們對於雜訊較不 敏感的緣故.此外,乏晰ID3學習演算法可以處理接近人類的思考方式的模 糊或是不精確的資料. 然而,這些學習演算法可能產生較複雜的(乏晰)分 類規則.在本論文中,我們提出一個新的方法,以綜合分析的技巧建構乏晰 決策樹並從而產生乏晰分類規則.在本論文中所提的方法可以得到較簡單 以及更佳的乏晰分類規則. The ID3 learning algorithm was proposed to solve the classification problems by Quinlan in 1983. It is a kind of batch learning strategy. Several extended ID3 learning algorithms combining the fuzzy set theory were proven to have better performances due to the fact that they are not sensitive to noises. Furthermore, fuzzy ID3 learning algorithms can deal with vague and imprecise data values associated with human thinking and perception. However, these learning algorithms may generate complexer (fuzzy) classification rules. In thisthesis, we propose a new method for generating fuzzy classification rules based on the construction of the fuzzy decision trees using compound analysis techniques to derive simpler and better fuzzy classification rules.