An Evolutionary RBFNN Learning Algorithm for Complex Classzification Problems

摘要

A self-optimizing approach for complex classifications is proposed in this paper to construct dynamical radial basis function neural network (RBFNN) models based on a specially designed genetic algorithm (GA). The algorithm adopts a matrix-form mixed encoding and specifically designed genetic operators to optimize the decayed-radius selected clustering (DRSC) process by co-evolving all of the parameters of the network's layout. The individual fitness is evaluated as a multi-objective optimization task and the weights between the hidden layer and the output layer are calculated by the pseudo-inverse algorithm. Experimental results on eight UCI datasets show that the GA-RBFNN can produce a higher accuracy of classification with a much simpler network structure and outperform those models of neural network based on other training methods.