An extended numerical analysis of an intuitionistic fuzzy classifier for imbalanced classes

Abstract

Recognizing relatively smaller classes (called imbalanced classes) from data is an important task both from a theoretical and practical points of view. In many real world problems smaller classes are usually more interesting from the user point of view but they are more difficult to obtain by a classifier. This paper, which is a continuation of our previous works, discusses a classifier that is based on Atanassovs intuitionistic fuzzy sets (A-IFSs, for short) and shows that it can be a good tool for recognizing imbalanced classes. Our considerations are illustrated on benchmark examples. Special attention is paid to a detailed behavior of the classifier proposed (different measures besides the general accuracy are examined). A simple cross validation method is applied (with 10 experiments). Results are compared with a fuzzy classifier reported to be good in the literature. Also the influence of the type granulation, symmetrical or asymmetrical, and of the number of intervals is considered.