GMDH neural network algorithm using the heuristic self-organization method and its application to the pattern identification problem

Abstract

In this paper, the GMDH (group method of data handling) neural network algorithm using the heuristic self-organization method is proposed. The structure of the GMDH neural network is organized automatically by using the the heuristic self-organization method used in the GMDH algorithm. Furthermore, the optimal neuron's structures are selected automatically so as to minimize the values of the prediction error criterion AIC (Akaike's information criterion) and the useless neurons are eliminated from the neural network. So the GMDH neural network has a good generalization ability.