A cooperative coevolution algorithm of RBFNN for classification

Abstract

This article presents a new learning algorithm, CO-RBFNN, for complex classifications, which attempts to construct the radial basis function neural network (RBFNN) models by using a cooperative coevolutionary algorithm (Co-CEA). The Co-CEA utilizes a divide-and-cooperative mechanism by which subpopulations are coevolved in separate populations of evolutionary algorithms executing in parallel. A modified K-means method is employed to divide the initial hidden nodes into modules that are represented as subpopulation of the Co-CEA. Collaborations among the modules are formed to obtain complete solutions. The algorithm adopts a matrix-form mixed encoding to represent the RBFNN hidden layer structure, the optimum of which is achieved by coevolving all parameters. Experimental results on eight UCI datasets illustrate that CO-RBFNN is able to produce a higher accuracy of classification with a much simpler network structure in fewer evolutionary trials when compared with other alternative standard algorithms. © Springer-Verlag Berlin Heidelberg 2007.