Logistic GMDH-Type Neural Network and its Application to Identification of X-Ray Film Characteristic Curve

Abstract

The logistic Group Method of Data Handing (GMDH)-type neural network identifying a complex nonlinear system we propose is automatically organized using heuristic self-organization that is basic to GMDH algorithm. In this neural network, structural parameters such as the number of layers, the number of neurons per layer, useful input variables, and optimum neuron architectures are automatically determined using a prediction error criterion defined as Akaike’s Information Criterion (AIC) to produce an optimum neural network architecture suiting the complexity of the nonlinear system. In applying this neural network to the identification problem of the X-ray film characteristic curve, we found that modeling with such a neural network is more accurate than applying multiple regression analysis, a conventional neural network, and GMDH algorithm.