Modifying weighted fuzzy subsethood-based rule models with fuzzy quantifiers

Abstract

The use of fuzzy quantifiers in linguistic fuzzy models helps to build fuzzy systems that use linguistic terms in a more natural way. Although several fuzzy quantification techniques have been developed, the application of the existing techniques seems very limited. This paper proposes an application of fuzzy quantification to replace crisp weights in subsethood-based fuzzy rule models. In addition to the concern that fuzzy models should have high accuracy rate, attention has also been taken to maintain the simplicity of the generated fuzzy model. The objective is to produce quantifier-based fuzzy models which are not only readable but also practically applicable. The quantifier based fuzzy model is then applied to classification tasks. The classification accuracies of fuzzy models that use crisp weights, continuous quantifiers, multi-valued quantifiers and two-valued quantifiers are compared. Experimental results show that the classification accuracy of the fuzzy model that uses continuous quantifiers is: 1) as good as the classification accuracy of the fuzzy models that use crisp weights, and 2) in most cases, better than fuzzy models that use multi-valued quantifiers or two-valued quantifiers.