A Genetic Tuned Fuzzy Classifier Based on Prototypes

Abstract

It is known that main drawbacks of the k-Nearest Neighbors classifier are related to the need for keeping all the training prototypes. Although there are several approaches capable to significantly reduce the size of the case base, they damage the classification accuracy. We propose a novel fuzzy approach that significantly reduces the prototypes base and also improves the classification accuracy. Its good performance is evidenced by an experimental study involving 20 prototype based classifiers and 30 databases, in which the proposal is the only approach placed among the top performers in both reduction and accuracy.