在順序型類別的分類是實務上很常見的問題，至今有許多專家學者提出針對順序型類別資料的分類器方法，包含許多統計常用的模型和近年來被廣為引用的資料探勘方法。但是，多數分類器所採用的統計模型必須滿足前提假設才能做配適，例如資料必須符合均質性、常態性和獨立性。另一方面，評估分類器所使用的績效評估指標也攸關到最後決定分類器的決策，不恰當的績效評估指標可能會導致最後選擇的分類器效果不佳。本文建議使用加權kappa係數來評估順序型分類器的績效，並利用實際的誤判成本計算出的非對稱權重矩陣做加權，既接近真實情況，又能考慮到預測類別和實際類別的一致性。本文也嘗試將統計常用的線性判別分析與資料探勘中的十個分類方法做比較，以找到分類績效較好的分類器。

There are many conventional methods to classify ordered data into some specific classes, including statistical methods and data mining. But the use of statistical methods should be based on the assumption of normality, independence and homogeneity. The thesis aims to compare classifiers built with linear discriminant analysis and ten methods of data mining, and found that the further is powerless. Moreover, the performance index used to compare classifiers is related to the precise of decision. An improper performance index may lead to wrong choice of classifiers. The performance index proposed in this thesis is improved from weighted Kappa and considered asymmetrically weighted cost matrix calculated by the cost of misclassification. The results show that the performance index proposed is more credible.

目錄
第一章 緒論 1
1.1 研究動機與目的 1
1.2 研究流程 3
1.3 論文架構 4
第二章 文獻探討 5
2.1 類神經網路（Artificial Neural Networks）5
 2.1.1 類神經網路之簡介 5
 2.1.2 倒傳遞類神經網路及演算法 7
2.2 決策樹（Decision Tree）8
 2.2.1 決策樹之簡介 8
 2.2.2 C4.5及其演算法 9
2.3 線性判別分析（Linear Discriminant Analysis）10
2.4 其他WEKA分類器	11
2.5 分類器的績效評估指標 14
第三章 研究方法 17
3.1 單峰模型（Unimodal Model）17
3.2 分類器的績效評估指標 20
第四章 模擬與實例分析 29
4.1 模擬資料 30
 4.1.1 n=10000 31
 4.1.2 n=5000 33
 4.1.3 n=3000 35
4.2 實例資料 37
 4.2.1 Car資料 38
 4.2.2 Nursery資料 40
 4.2.3 LEV資料 42
 4.2.4 SWD資料 43
第五章 研究結論與建議 46
5.1 結論	46
5.2 建議	47
參考文獻	48

表目錄
表4.1 n=10000時的kappa_new和kappa_old以及loss ratio 33
表4.2 n=5000時kappa_new和kappa_old以及loss ratio 35
表4.3 n=3000時kappa_new和kappa_old以及loss ratio 37
表4.4 Car資料下kappa_new和kappa_old以及loss ratio 39
表4.5 Nursery資料下kappa_new和kappa_old以及loss ratio 41
表4.6 LEV資料下kappa_new和kappa_old以及loss ratio 43
表4.7 SWD資料下kappa_new和kappa_old以及loss ratio 45

圖目錄
圖1.1 研究架構圖 3
圖2.1 處理單元間傳導訊息流程圖 6
圖2.2 倒傳遞神經網路架構 7
圖2.3 RBF類神經網路架構 12
圖4.1 模擬資料歸屬類別示意圖 31
圖4.2 n=10000各分類器的績效評估值 32
圖4.3 n=5000各分類器的績效評估值 34
圖4.4 n=3000各分類器的績效評估值 36
圖4.5 Car資料下各分類器的績效評估值 38
圖4.6 Nursery資料下各分類器的績效評估值 40
圖4.7 LEV資料下各分類器的績效評估值	42
圖4.8 SWD資料下各分類器的績效評估指標值 44

英文部分：
1.Ben-David, A. and Frank, E. (1992), Accuracy old machine learning models versus “hand crafted” expert systems – A credit scoring case study, Expert Systems with Applications, 36,  5264-5271.

2.Ben-David, A., Sterling, L. and Tran, T, (2009), Adding monotonicity to learning algorithms may impair their accuracy, Expert Systems with Applications, 36, 6627-6634.

3.Berry, M. J. A. and Linoff, G. (1997), Data Mining Techniques for Marketing Sale and Customer Support, Wiley Computer, New York, NY.

4.Bishop, C. M. (1995), Neural networks for pattern, Oxford: Clarendon Press.

5.Bohanec, M. and Rajkovic, V. (1990), Expert system for decision making, Sistemica 1, 145-157.

6.Cohen, J. (1960), A coefficient of agreement for nominal scales, Edu. and Psych. Meas., 20, 37-46.

7.Cohen, J. (1968), Weighted kappa: Nominal scale agreement with provision for scaled disagreement or partial credit, Psych. Bull., 70, 213-220.

8.Domingos, P., and Pazzani, M. (1997), On the optimality of the simple Bayesian classifier under zero-one loss, Machine Learnings, 29, 103-130.

9.Frank, E., Hall, M., Pfahringer, B (2003), Locally Weighted Naive Bayes, 19th Conference in Uncertainty in Artificial Intelligence, 249-256.

10.Graham, P. and Jackson, R. (1993), The analysis of ordinal agreement data: Beyond weighted kappa, Journal of Clinical Epidemiology, 46, 1055-1062.

11.Holte, R. C. (1993), Very simple classification rules perform well on most commonly used datasets, Machine Learning, 11, 63-91.

12.John, G. H. and Langley, P. (1995), Estimating Continuous Distributions in Bayesian Classifiers, Eleventh Conference on Uncertainty in Artificial Intelligence, San Mateo, 338-345.

13.Landwehr, N., Hall, M. and Frank E. (2005), Logistic Model Trees, Machine Learning, 95, 161-205.

14.Olave, M., Rajkovic, V. and Bohanec, M. (1989), An application for admission in   public school systems. Expert Systems in Public Administration, 145-160. 

15.Pinto da Costa, J. F., Alonso, H. and Cardoso, J. S. (2008), The unimodal model for the classification of ordinal data, Neural Networks, 21, 78–91.

16.Press, W., Flannery, B., Teukolsky, S. and Vetterling, W. (1992). Numerical recipes in C: The art of scientific computing, 2nd edition, Cambridge : Cambridge University Press.

17.Quinlan, J. R. (1993) C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers.

18.Venables, W. N. and Ripley, B. D. (2002), Modern applied statistics with S, 4th edition, New York : Springer.

19.Witten, I. H. and Frank, E. (2005), Data Mining: Practical Machine Learning Tools and Techniques, 2nd edition, San Francisco, Calif, : Elsevier/Morgan Kaufmann.


網路部分：
WEKA, http://www.cs.waikato.ac.nz/~ml/weka/index.html