利用模糊归属函数改善支持向量机improvementof-中央大学电机系.pdfVIP

利用模糊歸屬函數改善支持向量機 Improvement of Support Vector Machines Using Fuzzy Membership Function 韓明峰、鍾鴻源 * * Ming-Feng Han, Hung-Yuan Chung 國立中央大學電機工程系 Department of Electrical Engineering, National Central University * E-mail: hychung@.tw also assume that if the data sets are in an imprecise 摘要 situation, such as classes overlap, it can be 本文研究目的是藉由訓練樣本間的模糊性質 represented by fuzzy sets. Therefore, a fuzzy 採掘出有用的特徵，以改善支持向量機（Support membership is created according to the property of Vector Machines ；SVMs ）的性能。在理論建構上， classes overlap. In Support Vector Machines (SVMs), 本文假設訓練樣本分布為兩個高斯密度函數且具 a fuzzy-penalizing parameter is defined to balance 有類重疊（ Class Overlap ）現象，並利用類重疊分 both margin width and model complexity. Results 布建立模糊歸屬函數(Fuzzy Membership Function) show a powerful Fuzzy Support Vector Machines 在支持向量機的目標函數設計上，本文利用歸屬函 （FSVMs ）. In addition, each data point has 數重新定義一個模糊懲罰參數（ Fuzzy-penalizing individual and valuable parameter coefficient. In parameter ），以每個訓練樣本存有的不同貢獻度去 order to verify this classifier, we the proposed method 平衡間隔大小與訓練誤差，最後呈現一種新的且有 with traditional SVMs is compared in the experiment. 效率的模糊支持向量機（Fuzzy Support Vector It is shown that the proposed FSVMs are superior to Machines ；FSVMs ）。為了驗證這個分類器，本研 traditional SVMs in terms of both testing accuracy