Método de selección de atributos por clase

In many domains it is required to solve classification problems where the objects of study are described with a large number of features, many of which can be redundant and/or irrelevant. In order to improve the quality in classification it is necessary to eliminate this kind of features. Feature se...

Full description

Bibliographic Details
Author: BARBARA BERENICE PINEDA BAUTISTA
Format: master thesis
Status:Versión aceptada para publicación
Publication Date:2009
Country:México
Institution:Instituto Nacional de Astrofísica, Óptica y Electrónica
Repository:Repositorio Institucional del INAOE
Language:Spanish
OAI Identifier:oai:inaoe.repositorioinstitucional.mx:1009/432
Online Access:http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/432
Access Level:Open access
Keyword:info:eu-repo/classification/Aprendizaje supervisado/Supervised learning
info:eu-repo/classification/Reconocimiento de patrones/Pattern recognition
info:eu-repo/classification/Selección de características/Feature selection
info:eu-repo/classification/Clasificación de patrones./Pattern classification
info:eu-repo/classification/cti/1
info:eu-repo/classification/cti/12
info:eu-repo/classification/cti/1203
Description
Summary:In many domains it is required to solve classification problems where the objects of study are described with a large number of features, many of which can be redundant and/or irrelevant. In order to improve the quality in classification it is necessary to eliminate this kind of features. Feature selection has been widely used for the elimination of redundant and/or irrelevant features. There are two types for feature selection: 1. Feature selection for all classes. 2. Feature selection by class. Feature selection by class emerges with the idea that each class of a classification problem may have different properties and it should be described by a different feature subset. In this thesis a method of feature selection by class. The proposed method allows, by applying the one-against-all class binarization technique, the use of conventional feature selectors. Because supervised classifiers do not allow using a different feature subset for each class, it is also proposed to use a classifier ensemble and a new strategy decision for taking advantage of feature selection by class. The experimental results showed that in most cases the classification accuracy is improved when feature selection by class is used, compared against feature selection for all classes or without feature selection.