建立在位元映射索引上的特徵選取方法

Abstract

案例式推論是利用過去的案例和經驗來解決目前問題的技術。但是由於知識不容易被完整地呈現，使得如何找到足夠的特徵來充分地呈現案例特性成為一個關鍵的問題。在這篇論文中我們針對最佳解和近似最佳解，分別提出了以位元映射為基礎之特徵選取法及輔以辨識矩陣之位元映射特徵選取法。利用位元映射索引技術和約略集合的觀念，在現有的資料庫中自動選取重要的欄位。同時針對這二種方法，我們也提供了明確的定義及相對應的演算法。最後，經由和相關特徵選取技術的實驗比較結果，證明我們提出的特徵選取方法能以更快速的方式提供了正確的特徵選取結果。

CBR(Case-Based Reasoning) is a problem solving technique that reuses past cases and experiences to find a solution to current problems. A critical issue in case-based reasoning is to select the correct and enough features to represent a case. However, this task is difficult to carry out since such knowledge is often exhaustively captured and cannot be represented successfully. In this thesis, the new, efficient feature selection methods originated from bitmap indexing and rough set techniques will be proposed. There are two methods, including bitmap-based feature selection method and bitmap-based feature selection method with discernibility matrix, are proposed for discovering the optimal and nearly optimal feature sets for decision–making problems. And the corresponding indexing and selecting algorithms for such feature selection methods are also proposed. Finally, some experiments and comparisons are given and the result shows the efficiency and accuracy of our proposed methods.