The Problem of Fragile Feature Subset Preference in Feature Selection Methods and A Proposal of Algorithmic Workaround

0348710 - ÚTIA 2011 RIV TR eng C - Konferenční příspěvek (zahraniční konf.)
Somol, Petr - Grim, Jiří - Pudil, Pavel
The Problem of Fragile Feature Subset Preference in Feature Selection Methods and A Proposal of Algorithmic Workaround.
Proc. 2010 Int. Conf. on Pattern Recognition. Istanbul: IEEE Computer Society, 2010, s. 4396-4399. ISBN 978-0-7695-4109-9. ISSN 1051-4651.
[20th International Conference on Pattern Recognition. Istanbul (TR), 23.08.2010-26.08.2010]
Grant CEP: GA ČR GA102/07/1594; GA ČR GA102/08/0593; GA MŠMT 1M0572
Grant ostatní: GA MŠk(CZ) 2C06019
Výzkumný záměr: CEZ:AV0Z10750506
Klíčová slova: feature selection * machine learning * over-fitting * classification * feature weights * weighted features * feature acquisition cost
Kód oboru RIV: BD - Teorie informace
http://library.utia.cas.cz/separaty/2010/RO/somol-the problem of fragile feature subset preference in feature selection methods and a proposal of algorithmic workaround.pdf

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.
Trvalý link: http://hdl.handle.net/11104/0189152