本文目標在於介紹一個特徵委員會分類器的方法，處理兩不同比例族群的分類問題。第一、根據「一人一票，票票等值」的投票法則，將特徵委員會分類器中，每個特徵的分類結果投票計數，方能得到投票總數的判斷結果。因此，本文提供一個簡潔並且可以產生大量特徵的方法進行投票。第二、本文將提出一個估測族群比例的方法，藉此調動投票門檻值以達到更好的分類效果。由於各種疾病的盛行率可能非常的不同，所以這個方法對於醫學診斷而言非常重要。第三、本文所開發的特徵委員會分類器將用直觀的篩選標準，決定投票特徵候選人的成員數目。
為了論證本研究的方法，將選用睡眠腦波資料作為實驗資料，分成慢波睡眠期(slow wave sleep , SWS)與非慢波睡眠期(non-slow wave sleep stages, NSWS)兩族群類別。本研究選用離散小波轉換的訊號處理方法，將睡眠資料建立了總共1432個特徵作為投票特徵候選人。藉由本文提出的方法，估測非慢波及慢波睡眠期的族群比例值；可以根據此估測族群比例值，調動投票分類的投票門檻值。結果證實本研究的技術方法，可以於調整投票的投票門檻值後，使得分類的效果更好。

The goal of this work is to introduce a committee machine method for binary classification problems. The proposed approach has several distinct properties. First, based on the “one feature one voting” rule, the decision made by the committee machine is generated by counting the classification results produced by each of the features. Due to such simplicity, the proposed approach can easily be applied to problems with large number features. Second, by introducing a technique to estimate the class ratio, the proposed approach can adaptively adjust the voting threshold for better classification results. This is particularly important for medical diagnostic problems since the prevalence of diseases can be considerably different. Third, to determine the size of the committee machine, the proposed approach uses a simple rule to determine the number of features that can be included in the committee machine.
To demonstrate its effectiveness, the proposed approach was used to solve the slow wave sleep (SWS) detection problems by classifying the slow wave sleep (SWS) periods from the remaining non-slow wave sleep stages (NSWS). In specific, by using discrete wavelet transforms (DWT), this work generated 1432 features from the EEG signals. By using the proposed approach to solve a number of SWS detection problems with different NSWS/SWS ratios, the results show that the proposed techniques of adaptively determining the voting threshold and committee machine size can indeed improve the classification results.

[論文審定書+i]
[致謝+ii]
[摘要+iii]
[目錄+v]
[圖次+vii]
[表次+ix]
[第一章 緒論+1]
[1.1 前言+1]
[1.2 研究動機及目的+1]
[第二章 實驗資料處理與規劃+3]
[2.1 資料來源+3]
[2.1.1 睡眠檢查+3]
[2.1.2 睡眠分期+5]
[2.2 實驗前處理流程+8]
[2.2.1 族群類別區分+9]
[2.2.2 挑除不良訊號+10]
[2.2.3 建立樣本群+12]
[2.2.4 建立小波特徵委員會+14]
[第三章 分類流程與方法+18]
[3.1 分類方法+18]
[3.1.1 分類結果的各項指標+18]
[3.1.2 門檻值切割分類方法+20]
[3.1.3 二項分佈原理與投票分類方法+22]
[3.2 分類流程與核心方法 +24]
[3.2.1 投票成員篩選+25]
[3.2.2 族群比例估測方法+25]
[3.2.3 調動投票門檻值方法+27]
[第四章 實驗結果與討論+29]
[4.1 樣本群組成分析+29]
[4.2 族群比例估測結果+29]
[4.3 調動投票門檻值結果+32]
[4.4 投票分類結果+33]
[第五章 結論+41]
[參考文獻+43]
[附錄+45]

[1] Dhar V. Data science and prediction. Communication of the ACM. 2013;56 (12):64-73.
[2] Wu X, Zhu X, Wu GQ, Ding W. Data mining with big data. IEEE Trans. Knowl. Data Eng. 2014;26(1):97-107.
[3] Fayyad UM, Wierse A, Grinstein GG. Information visualization in data mining and knowledge discovery. 2002.
[4] Pal SK, Mitra P. Pattern recognition algorithms for data mining. 2004.
[5] 泛科技：大數據（Big Data）是什麼？
2016. URL: https://panx.asia/archives/43136
[6] Diebold FX. A personal perspective on the origin(s) and development of 「Big Data」: the phenomena, the term and the discipline.
[7] INSIDE：7個你不可不知的大數據定義。
2015. URL: https://www.inside.com.tw/2015/02/13/big-data-2-7-definitions-you-should-know-about
[8] Bello-Orgaz G, Jung JJ,Camacho D. Social big data: recent achievements and new challenges. 2016;28:45-59.
[9] Manyika J, Chui M, Brown B, Bughin J, Dobbs R, Roxburgh C, Byers AH. Big Data: The Next Frontier for Innovation, Competition, and Productivity. McKinsey Global Institute, 2011.
[10] Joachims T. Optimizing search engines using clickthrough data. In Proceedings of the ACM Conference on Knowledge Discovery and Data Mining(KDD), 2002.
[11] Lam L, Suen CY, Application of majority voting to pattern recognition: an analysis of its behavior and performance. IEEE Trans. Systems Man Cybernet. 1997;27(5):553-568.
[12] Bauer E, Kohavi R. An empirical comparison of voting classification algorithms: bagging, boosting, and variants. Machine Learning. 1999;36:105-139.
[13] Agarwal R, Gotman, J. Computer-Assisted sleep staging. IEEE Transactions on Biomedical Engineering. 2001;48:1412-1423.
[14] 蔡東遠，2010，使用不同組合的腦電圖、眼動圖及肌電圖訊號自動分類清醒期及淺睡期，碩士論文，國立中山大學機械與機電工程學系。
[15] The AASM manual for the scoring of sleep and associated events: rules, terminology, and technical specification.
[16] 陳世昌，2010，使用不同組合的腦電圖、眼動圖及肌電圖訊號自動偵測慢波睡眠，碩士論文，國立中山大學機械與機電工程學系。
[17] 李宜容，2010，使用不同組合的腦電圖、眼動圖及肌動圖訊號自動偵測快速動眼睡眠期，碩士論文，國立中山大學機械與機電工程學系。
[18] Kuncheva L, Whitaker C, Shipp C, Duin R. Limits on the majority vote accuracy in classifier fusion. Pattern Analysis and Applications. 2000.
[19] Demirekler M, Altıncay H. Plurality voting based multiple classifier systems: Statistically independent with respect to dependent classifier sets. Pattern Recogn. 2002;35(11):2365-2379.
[20] Ruta D, Gabrys B. A theoretical analysis of the limits of majority voting errors for multiple classifier systems. Pattern Analysis and Applications 2002;5 (4):333-350.
[21] Littlestone N, Warmuth MK. The weighted majority algorithm. Information and Computation. 1994;108:212-261.