本論文中以支撐向量機為基礎,使用規則擷取(Rule Extraction)的方式,配合連續覆蓋法(sequentialcovering algorithm)與模糊理論(Fuzzy logic)提出一個三階段的模糊法則建構演算法來達成支撐向量機器中的規則擷取,第一階段首先分析支撐向量機器的特徵向量與一般訓練資料的不同點,接著藉由空間分佈的概念去篩選出本論文所需要的特徵向量。第二階段利用特徵向量來建構出一般性的規則,這當中會針對所建構出來的規則進行覆蓋(covering)的過程,以達到規則數減少同時涵蓋率最大化的目標。第三階段則將規則轉換成模糊規則並以目標函數暨學習機制修正模糊規則。利用模糊理論中歸屬函數的優點,使得每個規則都有自己的歸屬函數,在判別類別時有了歸屬值可以比較,不再是單純的0、1,因此訓練資料在分類判斷上就可以依據歸屬程度來分類,降低誤判的機會。較於其他分類演算法,此演算法的優點在於所產生的法則具有法則數量較少以及保持高正確性。
In this thesis, a three phases of rule extraction algorithm, based on sequential covering algorithm and fuzzy logic, is proposed to extract fuzzy rules for support vector machines. In first phase, feature vectors are selected by analyzing the properties of training patterns. In second phase, rules are generated by using feature vector. Within this phase, sequential covering algorithm is used to reduce the rules and maximize the covering scope. In third phase, fuzzy logic is introduced so that general rules are transferred to fuzzy rules. Besides, learning mechanism is used to achieve optimization. Compared to other classification algorithms, our proposed algorithm achieves lower rules and high accuracy.