頁籤選單縮合
題 名 | 一個有效率的演算法用來設計k-NNR分類器=An Efficient Algorithm for the Design of a k-NNR Classifier |
---|---|
作 者 | 何信瑩; 楊智傑; | 書刊名 | 逢甲學報 |
卷 期 | 36 1999.12[民88.12] |
頁 次 | 頁53-72 |
分類號 | 310.153 |
關鍵詞 | k-NNR法則; Pareto解集合; IRIS資料; Heart資料; k-NNR; Pareto solution; IRIS data; Heart data; |
語 文 | 中文(Chinese) |
中文摘要 | 本論文提出一個有效率的演算法用設計選擇型k-NNR分類器,本演算法能夠從訓 練集中選擇出一組集合數目最少且分類正確率最高的子集合作為分類器的參考集,而且還能 提供一組同時考量參考集數目和分類正確率的Pareto解集合。本演算法對於參考集樣本的選 擇是以訓練集中每個樣本對於分類正確率貢獻代表性的排名為依據,這些排名的結果是經過 一個系統化推理的過程所得到的,而且不需重複的步驟和啟發式的知識。我們採用IRIS資料 和一組醫學資料heart作為實驗的分類樣本,實驗結果顯示本演算法比目前相關研究中成果 最好的基因演算法可以更快速有效地獲得一個分類正確率較高,且集合數目較少的參考集。 |
英文摘要 | In this paper an efficient algorithm for the design of a selected k-nearest neighbor (k-NN) classifier based on the concepts of an optimal subset selection from a given training set is presented. The algorithm can lead to both better computational efficiency and higher classification accuracy than the existent methods. Furthermore, it can give a Pareto solution for the cardinality of reference set and the classification accuracy. The samples are selected in order based on the ranks of significance of their contribution for the classification accuracy using a systematic reasoning method without iterations and heuristic knowledge. An experimental study with the Fisher's IRIS data set and a medical heart data set has been carried out. Experimental results show that the smaller subsets and higher classification accuracy can be obtained fast with linear time complexity than the genetic algorithm approaches which have been used to solve the editing k-NN reference set problems. |
本系統中英文摘要資訊取自各篇刊載內容。