Number of the records: 1
The Problem of Fragile Feature Subset Preference in Feature Selection Methods and A Proposal of Algorithmic Workaround
- 1.
SYSNO ASEP 0348710 Document Type C - Proceedings Paper (int. conf.) R&D Document Type Conference Paper Title The Problem of Fragile Feature Subset Preference in Feature Selection Methods and A Proposal of Algorithmic Workaround Author(s) Somol, Petr (UTIA-B) RID
Grim, Jiří (UTIA-B) RID, ORCID
Pudil, Pavel (UTIA-B) RIDSource Title Proc. 2010 Int. Conf. on Pattern Recognition. - Istanbul : IEEE Computer Society, 2010 - ISSN 1051-4651 - ISBN 978-0-7695-4109-9 Pages s. 4396-4399 Number of pages 4 s. Publication form flash - flash Action 20th International Conference on Pattern Recognition Event date 23.08.2010-26.08.2010 VEvent location Istanbul Country TR - Turkey Event type WRD Language eng - English Country TR - Turkey Keywords feature selection ; machine learning ; over-fitting ; classification ; feature weights ; weighted features ; feature acquisition cost Subject RIV BD - Theory of Information R&D Projects GA102/07/1594 GA ČR - Czech Science Foundation (CSF) GA102/08/0593 GA ČR - Czech Science Foundation (CSF) 1M0572 GA MŠMT - Ministry of Education, Youth and Sports (MEYS) CEZ AV0Z10750506 - UTIA-B (2005-2011) Annotation We point out a problem inherent in the optimization scheme of many popular feature selection methods. It follows from the implicit assumption that higher feature selection criterion value always indicates more preferable subset even if the value difference is marginal. This assumption ignores the reliability issues of particular feature preferences, overfitting and feature acquisition cost. We propose an algorithmic extension applicable to many standard feature selection methods allowing better control over feature subset preference. We show experimentally that the proposed mechanism is capable of reducing the size of selected subsets as well as improving classifier generalization. Workplace Institute of Information Theory and Automation Contact Markéta Votavová, votavova@utia.cas.cz, Tel.: 266 052 201. Year of Publishing 2011
Number of the records: 1