Logistic GMDH-type neural networks and their application to the identification of the X-ray film characteristic curve

Abstract

Logistic group method of data handing (GMDH)-type neural networks identifying a complex nonlinear system are proposed. Logistic GMDH-type neural networks are automatically organized by using the heuristic self-organization method which is used in the GMDH method. In the logistic GMDH-type neural networks, the structural parameters such as the number of layers, the number of neurons in each layer, useful input variables and optimum neuron architectures are automatically determined by using the error criterion derived from the AIC (Akaike's Information Criterion). This way, optimum neural network architectures which fit the complexity of the nonlinear system are produced. The logistic GMDH-type neural networks have been applied to the identification problem of the X-ray film characteristic curve. It has been found that the modeling with the logistic GMDH-type neural networks is more accurate than when multiple regression analysis, the conventional neural networks and the GMDH method are used.