Desarrollo de clasificadores basados en reglas de asociación de clase

Classification based on Class Association Rules (CARs) or associative classification is a data mining technique that consists of, given a training instance set, finding certain characteristics in the instances in order to build rules that are subsequently used for classifying unseen instances. Assoc...

Descripción completa

Detalles Bibliográficos
Autor: RAUDEL HERNANDEZ LEON
Tipo de recurso: tesis doctoral
Estado:Versión aceptada para publicación
Fecha de publicación:2011
País:México
Institución:Instituto Nacional de Astrofísica, Óptica y Electrónica
Repositorio:Repositorio Institucional del INAOE
Idioma:español
OAI Identifier:oai:inaoe.repositorioinstitucional.mx:1009/688
Acceso en línea:http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/688
Access Level:acceso abierto
Palabra clave:info:eu-repo/classification/Minería de datos/Data mining
info:eu-repo/classification/Reconocimiento de patrones/Pattern recognition
info:eu-repo/classification/Clasificación/Classification
info:eu-repo/classification/cti/1
info:eu-repo/classification/cti/12
info:eu-repo/classification/cti/1203
Descripción
Sumario:Classification based on Class Association Rules (CARs) or associative classification is a data mining technique that consists of, given a training instance set, finding certain characteristics in the instances in order to build rules that are subsequently used for classifying unseen instances. Associative classification has been used in different tasks, for example: text classification, text segmentation, and automatic image annotation, among others. However, associative classification methods still have some weaknesses. In this doctoral dissertation we propose an algorithm called CAR-CA, which introduces a new pruning strategy that allows to obtain specific rules with high values of the quality measure. Besides, we introduce two classifiers based on CARs, CAR-IC and CAR-NF, both use a new way for ordering the set of CARs based on the rule size, a new covering criterion that considers the inexact coverage when any rule covers the new instance, and a new strategy for deciding the class of a new instance. Additionally, these classifiers use as threshold for the quality measure, the minimum value that avoids ambiguity at the classification stage. In particular, The CAR-NF classifier introduces the use of the Netconf measure to compute the set of CARs. The experimental results show that the proposed CARs based classifiers CAR-IC and CAR-NF have better performance than the main successful classifiers based on CARs.