Revised Margin-Maximization Method for Fuzzy Nearest Prototype Classification

In this paper, we study the nearest prototype classification, which is a classification task using labeled prototypes. In previous work, we have revised our margin-maximization method for the nearest prototype classification. Its optimization problem is formulated using DC (Difference of Convex) functions and solved using CCP (Convex-Concave Procedure), which is a k-means-like algorithm. In this paper, we apply the revised method for fuzzy nearest prototype classification, in which the function selecting the closest prototype is fuzzified, and a label of a given instance is predicted using more than one nearest prototype. Through a numerical study, we examine the characteristics of the proposed method.