In many real problems, it is important to evaluate the precision ofclassifiers, as well as analyze the knowledge of this classifier to discover newknowledge or to offer some explanation of the given classification. Symbolicclassifiers are indicated when knowledge acquisition is needed, but, in somedomains, they show precision rates lower than precision rates of classifiers inducedby other types of learning algorithms. One solution is to evolve symbolicclassifiers into only one symbolic classifier. In this work we propose GAESCSG,a genetic algorithm that evolves symbolic classifiers into a single classifier.This algorithm uses crossover and mutation operators and also uses specializationand generalization operators. The experiments realized with this algorithmshows interesting results, because GAESC-SG shows more stable resultson number of rules and generations, related to GAESC.
http://www.lbd.dcc.ufmg.br/colecoes/enia/2012/0024.pdfCaso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web