Universidad de Burgos Repositorio Repositorio
Por favor, use este identificador para citar o enlazar este ítem: http://hdl.handle.net/10259/4814

Ver estadísticas de uso
Título : On feature selection protocols for very low-sample-size data
Autor : Kuncheva, Ludmila I. .
Rodríguez Diez, Juan José
Publicado en: Pattern Recognition. 2018, V. 81, p. 660-673
Editorial : Elsevier
Fecha de publicación : sep-2018
Fecha de disponibilidad: sep-2020
ISSN : 0031-3203
DOI: 10.1016/j.patcog.2018.03.012
Resumen : High-dimensional data with very few instances are typical in many application domains. Selecting a highly discriminative subset of the original features is often the main interest of the end user. The widely-used feature selection protocol for such type of data consists of two steps. First, features are selected from the data (possibly through cross-validation), and, second, a cross-validation protocol is applied to test a classifier using the selected features. The selected feature set and the testing accuracy are then returned to the user. For the lack of a better option, the same low-sample-size dataset is used in both steps. Questioning the validity of this protocol, we carried out an experiment using 24 high-dimensional datasets, three feature selection methods and five classifier models. We found that the accuracy returned by the above protocol is heavily biased, and therefore propose an alternative protocol which avoids the contamination by including both steps in a single cross-validation loop. Statistical tests verify that the classification accuracy returned by the proper protocol is significantly closer to the true accuracy (estimated from an independent testing set) compared to that returned by the currently favoured protocol.
Palabras clave: Feature selection
Wide datasets
Experimental protoco
Training/testing
Cross-validation
Licencia: http://creativecommons.org/licenses/by-nc-nd/4.0/
URI : http://hdl.handle.net/10259/4814
Versión del editor: https://doi.org/10.1016/j.patcog.2018.03.012
Aparece en las colecciones: Artículos ADMIRABLE

Ficheros en este ítem:

Fichero Descripción Tamaño Formato
Kuncheva-PR-2018.pdf3,53 MBAdobe PDFVisualizar/Abrir

Este ítem está sujeto a una licencia Creative Commons Licencia Creative Commons
Creative Commons

Los ítems del Repositorio Institucional de la Universidad de Burgos están protegidos por copyright, con todos los derechos reservados, a menos que se indique lo contrario.

 

Valid XHTML 1.0! DSpace Software Copyright © 2002-2008 MIT and Hewlett-Packard - Sobre DSpace