本論文提出一種有效率之混合式特徵子集選取方法,以克服因資料具多特徵所增加的計算數量,並且在分類法問題上獲得良好之學習能力。所提出之方法包含兩個步驟:(1)擬定一評估程序,用來產生特徵重要性之排序;(2)建構一新的二元式搜尋特徵子集(BSFS, Binary Search Feature Subset)演算法,以產生最佳特徵子集合。本篇並運用所提出之方法應用在模式感知網路(MPNs, Modular Perception Networks),並以實際之資料集來執行學習及運用。從實驗結果得知,輸入資料之特徵可減少約75%~88%,資料之計算量也可減少約67%~91%,並且可獲得一較小規模之模式感知網路(MPNs),其學習及測試能力保持和之前一些常用方法一樣,具有良好的水準。
In this paper, a simple and efficient feature subset selection method is proposed to overcome the curse of dimensionality and to obtain good learning performance on classification problems. The proposed method includes two steps: (1) Scheming an evaluation function to create the rank of the feature significance; (2) Constructing a new Binary Search Feature Subset (BSFS) algorithm to generate the optimum feature subset. In this study, the proposed method is applied on a Modular Perception Networks (MPNs) to learn the real word datasets. It shows that from the experimental results the feature of the input data can be decreased largely (less 75%~88%), the data presentations are reduced (less 67%~91%) and a small size MPNs can be procured with learning and testing performance maintained as the good level as before.