Chinese Journal of Computers   Full Text
Title	An Novel Ensemble Method of Feature Gene Selection Based on Recursive Partition-Tree
Authors	LI Xia1)，2) ZHANG Tian-Wen1) GUO Zheng1)，2)
Address	1)（Department of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001) 2)(Department of Biomedical Engineering and Bioinformatics, Harbin Medical University, Harbin 150086）
Year	2004
Issue	No.5(675-682)
Abstract &   Background	To identify disease genes from these gene expression profiles is critically important for disease, such as cancer, subtype discovery, diagnosis and pathology study. This paper proposes a feature gene selection method named EFST（Ensemble Feature Selection Based on Recursive Partition-Tree） which can be applied to select multiple feature gene groups from one training sample set，and defines a significance and stability measure for each selected feature in a way similar to the ensemble decision method of supervised machine learning. Authors apply the EFST method to analyse the published 2,000 gene expression profile data. The results indicate that the EFST method can be used not only to select feature genes and reduce the dimension of feature space, but also to increase significantly the disease prediction accuracy of many classification methods including SVM, nearest neighbor classifier, Fisher linear and Logistic nonlinear discriminant analysis. keywords gene expression profile; recursive partition-tree; feature selection; ensemble decision