整合式規則歸納系統

Abstract

分類問題的目的是由資料中歸納出一組分類規則，以表示屬性與類別 之間的關係，而這樣的規則對於專家系統的建立是非常重要的。到目前為 止，已有不少機器學習的方法被用來解決分類問題，例如資訊理論，模糊 邏輯，遺傳演算法，與類神經網路等，然而這些方法皆有其缺點或限制。 本篇論文整合資訊理論，模糊邏輯，與演化式計算等方法，提出一個兼具 效果與效率的整合式規則歸納策略，能自動求出分類問題中的模糊集合個 數，歸屬函數，及分類規則。同時也根據此策略實作一個視窗介面的整合 式規則歸納系統，並以鳶尾花及乳房瘤兩種分類問題進行實驗與比較。 The purpose of a classification problem is to induce a set of classification rules from data to characterize the dependence of classes on attributes. Moreover, these rules are of great importance on the development of expert systems. So far, many machine learning methods, such as information theory, fuzzy logic, genetic algorithms, and artificial neural networks, have been used to solve classification problems, but they all have disadvantages and limits. In this thesis, information theory, fuzzy logic, and evolutionary computation are integrated to propose the Integrated Rule Induction Strategy with both effectiveness and efficiency to derive automatically the number of fuzzy sets, the membership functions, and the classification rules. Furthermore, this strategy is implemented to develop IRIS (Integrated Rule Induction System), a windows-interface program, and the iris flower and breast cancer domain are used to make experiments and comparisons.