Interpretability and accuracy issues in evolutionary multi-objective fuzzy classifiers

Abstract

Fuzzy systems have remarkable capability to deal with imprecise and uncertain information existing in the real world complex problems. Evolutionary approaches, i.e., genetic algorithms are utilised to improvise the designing of fuzzy systems. During the design of fuzzy systems, interpretability and accuracy features are considered as an effort toward the improvement of performance and usability. One can only be improved at the cost of the other, leading to a new trade-off called interpretability-accuracy trade-off. On the other hand, the use of interval type-2 fuzzy sets in the development of fuzzy classifier is another dimension of improving it. In this paper, a fuzzy classifier named Engineering Student-Fuzzy Classification System (ES-FCS) is proposed and implemented using type-1 and type-2 fuzzy logic. The accuracy improvement has been studied by the application of linguistic hedges. The interpretability and accuracy assessment and their trade-off are experimentally studied in evolutionary multi-objective framework with both type-1 and type-2 fuzzy sets.