Presentation | 2005/11/11 A Selection Criterion for Robust Classifiers by Considering the Variance of Test Performance Ikumi SUZUKI, Shigeyuki OBA, Junichiro HIRAYAMA, shin ISHII, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In classification problems with gene expression profiles emerged from clinical field, we usually compare performances of multiple classifiers and select the best classifier by standard cross-validation technique. Reliability of the selected classifier is, however, often low due to the large expected variance of test performance caused by the small sample size compared to the high-dimensionality of gene expression data. In this study, we propose a new method, the Parametric Noise Bootstrap and Percentile (PNBP) method which selects a classifier that tends to select poor classifier avoiding a high risk of resulting in a poor performance. The PNBP method evaluates the risk of selecting poor clasifier by applying cross-validation to artificial datasets those are created by a bootstrap method. We applied the PNBP method to determine the number of input gene in Weighted-Voting (WV) classification method and examined the basic property of the PNBP for real gene expression data. The result showed that the PNBP method select a large number of input gene when the sample size is small, in contrast that the standard method often select an extremely and unrealistically small number. This suggests that our PNBP criterion is robust for the variance of test performance evaluated by cross-validation and works as a risk-avoiding for model selection criterion. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Microarray / Classifier Performance Estimation / Bootstrap / model selection |
Paper # | NLP2005-66,NC2005-58 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2005/11/11(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Neurocomputing (NC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Selection Criterion for Robust Classifiers by Considering the Variance of Test Performance |
Sub Title (in English) | |
Keyword(1) | Microarray |
Keyword(2) | Classifier Performance Estimation |
Keyword(3) | Bootstrap |
Keyword(4) | model selection |
1st Author's Name | Ikumi SUZUKI |
1st Author's Affiliation | Graduate School of Information Science Nara, Institute of Science and Technology() |
2nd Author's Name | Shigeyuki OBA |
2nd Author's Affiliation | Graduate School of Information Science Nara, Institute of Science and Technology |
3rd Author's Name | Junichiro HIRAYAMA |
3rd Author's Affiliation | Graduate School of Information Science Nara, Institute of Science and Technology |
4th Author's Name | shin ISHII |
4th Author's Affiliation | Graduate School of Information Science Nara, Institute of Science and Technology |
Date | 2005/11/11 |
Paper # | NLP2005-66,NC2005-58 |
Volume (vol) | vol.105 |
Number (no) | 418 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |