Ensembles de sistemas de classificação fuzzy intervalares: o papel dos índices de overlap com tomada de decisão baseada em funções de penalidade

  • Sidnei da Fonseca Pereira Jr. Computational Modeling PhD Program, Federal University of Rio Grande, Campus Carreiros, Rio Grande
  • Graçaliz Pereira Dimuro Center for Computational Science, Federal University of Rio Grande, Campus Carreiros, Rio Grande & Departamento de Estadística, Informática y Matemáticas, Universidad Pública de Navarra, Campus Arrosadia, Pamplona
  • Eduardo Nunes Borges Center for Computational Science, Federal University of Rio Grande, Campus Carreiros, Rio Grande
  • Tiago da Cruz Asmus Departamento de Estadística, Informática y Matemáticas, Universidad Pública de Navarra, Campus Arrosadia, Pamplona
  • Leonardo Ramos Emmendorfer Center for Computational Science, Federal University of Rio Grande, Campus Carreiros, Rio Grande
  • Giancarlo Lucca Center for Computational Science, Federal University of Rio Grande, Campus Carreiros, Rio Grande
  • Humberto Bustince Departamento de Estadística, Informática y Matemáticas, Universidad Pública de Navarra, Campus Arrosadia, Pamplona
Keywords: Interval-valued fuzzy sets, Interval-valued overlap functions, Penalty functions

Abstract

Fuzzy modeling is frequently used to deal with the problems involving approximate reasoning, such as classification problems. However, fuzzy membership functions defined in terms of real functions sometimes can not reflect the uncertainty of the domain specialists. Also, when considering the use of fuzzy quantities, we are executing operations through sets defined by real numbers. In this case, the propagation of errors can become important and affect the end result. This problem finds its solution within the concept of interval-valued fuzzy sets. This paper presents a fuzzy reasoning mechanism to be used in interval-valued fuzzy rule based classification systems (IVFRBCSs). For that, we consider different interval-valued overlap indices, constructed using interval-valued overlap functions, developing confidence and support measures, which are generally used to evaluate the degree of certainty or interest of a given association rule. By considering several interval-valued overlap indices, we obtain an ensemble of IVFRBCS. Then penalty functions are used as a consensus method for the decision-making related to the selection of the best class.
Published
2021-10-20
Section
Articles