An Efficient and Effective Microarray Feature Selection Scheme with Change Point Detection.
| Autor: | HUI-YI CHANG, 張蕙憶 |
|---|---|
| Rok vydání: | 2013 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 101 In recent years, the microarray data is a very effective medical diagnostic tool and it has been widely used to analyze the correlation between genes and disease. Because the microarray data has high dimension and small sample size, it will produce overfitting and very time-consuming. Therefore, a good feature selection method for gene relevant for sample classification is needed in order to improve computation time and accurary. Based on some literature, feature selection methods are divided into filters and wrappers. Thus, according to this structure, SVMSC (a novel kernel-based gene selection and classification scheme) and CCP (coefficient change point feature selection method) are proposed in this paper, respectively. SVMSC is not a feature selection method, but also an integrated approach of feature selection and classification. SVMSC use a variable importance measure from RBF kernel function for selecting genes. This kernel function will also be used in the following SVM classifier. Experiment results show that most of the best performances are associated with SVMSC. CCP is based on Glmnet (Generalized Linear Models via Elastic-Net). Using the number of variables as a standard for selection of α. Then, by change point detection, select a very important variable effectively. Experiment results show that CCP has not only high enought accurary but also save most of computation time. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|