Differential evolution based multiple vector prototype classifier

Abstract

In this article we introduce differential evolution based multiple vector prototype classifier (shortly MVDE). In this method we extend the previous DE classifier so that it can handle several class vectors in one class. Classification problems which are so complex that they are simply not separable by using distance based algorithms e.g. differential evolution (DE) classifier or support vector machine (SVM) classifier have troubled researchers for years. In this article, we propose a solution for one area of this problem type in which we extend DE classifier in a way that we allow several class vectors to exist for optimizing one class. This way a part of such complex data can be handled by one vector and other part can be handled by another vector. Differential evolution algorithm is a clear choice for handling such a multiple vector classification tasks because of its remarkable optimization capabilities. MVDE classifier is tested with several different benchmark classification problems to show its capabilities and its performance is compared to DE classifier, SVM and backpropagation neural network classifier. MVDE classifier managed to get best classification performance of these classifiers and clearly indicates it has a potential in this type of classification problems.