Complex-valued support vector machines based on multi-valued neurons

Abstract

In this paper, we propose complex-valued support vector machines (CVSVMs) which are a new type of support vector machines (SVMs) based on multi-valued neurons (MVNs). An MVN which is a type of complex-valued neurons is a component of the proposed CVSVM. The features of the proposed CVSVM are: 1) it has a multi-valued complex output; 2) it provides the generalization ability by a decision boundary with the maximal margin; 3) it can deal with non-linear classification by using a kernel function. Experimental results for some famous benchmark problems show the effectiveness of the proposed CVSVM.