本研究之目的在於提出一個植基於合成屬性(integrated feature)的二元分類法，其主要是先利用「主成份分析」法將原本互為相關「屬性變數」的「屬性向量」轉化為任兩個「主成份」皆不相關的「主成份向量」，目的在於消除「屬性變數」之間的相關性太高時導致的多元共線性（multi-collinearity）問題。
將「屬性向量」轉化為「主成份向量」後，依據「主成份」與類別變數(label)之間相關係數 (correlation coefficient)的平方進行「主成份」排序，再逐步將相關係數top 1、top 2、top 3 ... 的主成份加入資料集中，本方法將自動探索並利用合成屬性與類別變數之間的關係，來提升分類的正確性。合成屬性是「主成份」與權重向量的線性組合，在給定的訓練樣本下，我們決定每個「主成份」的權重，使得依據合成屬性分類的準確度(Accuracy)最大化，由於計算最佳權重向量的困難，於是我們先利用線性判別分析(Linear Discrimination Analysis, LDA)演算法取得優化的權重向量，再以基因演算法(Genetic Algorithm)來逼近最佳的權重向量與分類的臨界值。本研究以醫學研究資料集進行實驗，實驗結果顯示本研究所提之二元分類器，在訓練樣本分類準確度可達到99.29%，測試樣本分類準確度則可達到97.18%。

Binary classification method predicts the class of an object based on the associated feature vector. The main logic behind the classification is to regress the feature vector on the class variable. Due to the high correlation between feature variables, we apply principal component analysis to obtain the uncorrelated transformed features to mitigate the multicollinearity problem.For maintaining the simplicity and flexibility of modeling the relationship between the feature vector and the class variable, we propose to predict the class of an object based on an integrated feature which is the linear combination of the features and the powers of each individual feature.The weights on the original features and their powers, and the corresponding classification threshold are determined by the genetic algorithm to maximize the classification accuracy based on the training dataset.To speed up the convergence rate of the genetic algorithm, the solution obtained from the linear discrimination analysis is used as the starting point of the genetic algorithm.We apply the proposed procedure on the benchmark datasets, and experimental results demonstrate the effectiveness of our algorithm.

[1] Olli Lahdenoja, Mika Laiho, and Ari Paasio,”Reducing the feature vector length in local binary pattern based face recognition,” IEEE International Conference on Image Processing, Genoa, Italy, Sept. 11-14, 2005, Vol. 2, pp. 914-917.

[2] Timo Ojala, Matti Pietikainen, and David Harwood, ”A comparative study of texture measures with classification based on feature distributions,” Pattern Recognition, Vol. 29, No. 1, pp. 51-59, 1996.

[3] Lichman, M. (2013). UCI Machine Learning Repository [http://
archive.ics.uci.edu/ml]. Irvine, CA: University of California, School
of Information and Computer Science.

[4] S. Aruna et al. (2011), Knowledge based analysis of various statistical tools in detecting breast cancer.

[5]Fisher, R.: The use of multiple measurements in taxonomic problems. Annals of Eugenics 7(7), 179_188 (1936)

[6]Izenman A.J. (2013) Linear Discriminant Analysis.In: Modern Multivariate Statistical Techniques.Springer Texts in Statistics. Springer, New York, NY

[7] D.E. Goldberg, Genetic Algorithms in Search, Optimisation, and Machine Learning, Addison Wesley (1989)

[8] Pearson, K. (1901). ”On Lines and Planes of Closest Fit to Systems
of Points in Space”. Philosophical Magazine. 2 (11): 559-572.

[9] Vellore, Tamil Nadu,India “ A SURVEY OF DIMENSIONALITY REDUCTION AND CLASSIFICATION METHODS” , International Journal of Computer Science & Engineering Survey (IJCSES) Vol.3, No.3, June 2012

[10] Zhao, Z., Wang, L., Liu, H. and Ye, J. (2013). On Similarity Preserving Feature Selection.IEEE Transactions on Knowledge and Data
Engineering, 25(3), pp.619-632.