最近鄰居分類演算法的研究及應用

Study and Application of NNDR Algorithm

鍾日富(Jih-Fu Chung)

指導教授 ： 廖炯州

共同指導教授 ： 葉雲奇(Yun-Chi Yeh)

健行科技大學/電資學院/電子工程系碩士班/碩士(2020年)

摘要

本篇論文提出最近鄰居分類演算法(nearest neighbor decision rule, NNDR)的研究及應用，它是以群聚分析(Cluster Analysis)演算法為理論基礎，並將之應用於辨識歷屆畢業校友的問卷為範例說明。本範例之目的是透過問卷調查，瞭解本系所設計之課程內容是否合乎就業技能之所需。辨識過程如下：(1)以問卷方式詢問本系的畢業系友，他們對於本系的課程設計內容與就業所需兩者之間的相關程度，本範例將相關程度分成 (a)極為相關、(b)高度相關、(c)中度相關、(d)低度相關、及(e)少有相關等5種程度表示；(2)依據此相關性的調查結果，以最近鄰居分類演算法(NNDR)辨識本系的課程設計內容是否合乎就業所需。亦即判斷本系的課程設計內容是否還需要再作調整。本範例將課程設計內容分成(a)符合、(b)不足、(c)缺乏、(d)不實用、及(e)無用等5種調整程度表示。其中，最近鄰居分類演算法(NNDR)的辨識，分成如下的三個步驟：(i)計算五種不同調整程度之加權差異度的中心值；(ii)計算待辨識的加權差異度它與五種課程調整程度之間的歐式距離(Euclidean distance)；(iii)決定課程的調整程度。本篇論文經過多次的實驗，確認本文提出的最近鄰居分類演算法是ㄧ個簡單有效的辨識方法，正確辨識率達99%以上。

關鍵字

最近鄰居分類法 ； 教學評量 ； 群聚分析

並列摘要

This study proposes the research and application of the nearest neighbor decision rule (NNDR), which is based on the cluster analysis algorithm and is applied to a questionnaire to identify previous graduates as an example for description. The purpose of this example is to find out whether the content of the courses designed by the department of Electronic Engineering, ChienHsin University of Science and Technology, is suitable for employment skills through questionnaires. The identification process is as follows: (1) Ask graduate students in the department by questionnaires about the degree of correlation between the curriculum design content and employment requirements of the department. This example divides the degree of correlation into (a) extremely relevant, (b) highly relevant, (c) Moderatelyrelevant, (d) lowly relevant, and (e) few relevant of correlation (NNDR). Identify whether the curriculum design content of this department is suitable for employment.(2) Based on the results of this correlation survey, use the nearest neighbor algorithm (NNDR) to identify whether the curriculum design content of this department is suitable for employment. That is, whether the curriculum designs content of this department needs to be adjusted again. This example divides the curriculum design content into five adjustment levels: (a) conformity, (b) deficiency, (c) lack, (d) impractical, and (e) useless. Among them, the identification of the nearest neighbor separation algorithm (NNDR) is divided into three steps as follows: (i) calculate the center value of the weighted difference of five different adjustment degrees; (ii) calculate the weighted difference to be identified. The Euclidean distance between the five course adjustment levels; (iii) determines the degree of course adjustment. The results ofmany experiments show that the proposed nearest-neighbor separation algorithm is a simple and effective identification method. The correct recognition rate is over 99%.

並列關鍵字

The nearest neighbor decision rule ； teaching evaluation ； cluster analysis

參考文獻

[1] Y.C. Yeh, H.J. Lin, and W.J. Wang, Classification of ECG signals by using a Cluster Analysis method, Biomedical Engineering Society 2008 Annual Symposium, 2008 (生物醫學工程科技研討會), Chang Gung University, Taipei, December 12, 2008.

[2] 鍾日富, 葉雲奇, 廖炯州, 人工智慧的應用：以群聚分析為基礎的心電圖訊號辨識器, 台灣機電工程國際學會第四屆全國學術研討會,高雄,台灣, February 15-16, 2018.

[3] E. Avci, and Z.H. Akpolat, Speech recognition using a wavelet packet adaptive network based fuzzy inference system, Expert System with Application, vol. 31, pp. 495-503, 2006.

[4] K.K. Seo, An application of one-class support vector machines in content-based image retrieval, Expert System with Application, vol. 33, pp. 491-498, 2007.

[5] C.H. Wang, Recognition of semiconductor defect patterns using spatial filtering and spectral clustering, Expert System with Application, vol. 34, pp. 1914-1923, 2008