Feature selection based on correlations of gene-expression values for cancer prediction
Conference proceedings article
Authors/Editors
Strategic Research Themes
No matching items found.
Publication Details
Author list: Pattanateepapon A., Suwansantisuk W., Kumhom P.
Publisher: Hindawi
Publication year: 2016
Start page: 716
End page: 721
Number of pages: 6
ISBN: 9781467377911
ISSN: 0146-9428
eISSN: 1745-4557
Languages: English-Great Britain (EN-GB)
Abstract
Microarray-based cancer prediction can save lives, but is difficult to perform accurately due to noise, outliers, and inherent anomalies in gene-expression measurements. One important method to improve accuracy, sensitivity, and specificity of cancer prediction is feature selection, where only relevant genes participate in classification. Current features-selection methods normally have high computational-cost or require trial and error to find a suitable parameter. We develop a feature-selection method called the support-vector-machine recursive feature-elimination (SVMRFE) with minimum Pearson correlation. This method selects genes that together classify well the samples and are weakly correlated, meaning that the selected genes are both relevant for classification and small in number. The proposed method preprocesses microarray data and adjusts automatically and adaptively a parameter to produce high classification performance. We compare the proposed method with some existing ones in terms of accuracy, sensitivity, and specificity. The results show that the proposed method outperforms and ties with a close competitor, the SVMRFE with maximum relevance and minimum redundancy, in 45 out of 72 test cases. In addition, the proposed method outperforms other four classic baselines in 3 out of 6 test cases. The proposed method of feature selection is easy to implement, systematic, and has practical applications to bioinformatics. ฉ 2016 IEEE.
Keywords
minimum redundancy and maximum relevance
Contact Information