Revised GMDH-type neural networks using PSS criterion
Tadashi Kondo · 2004
In this paper, a revised GMDH-type neural networks algorithm using PSS criterion for model selection is proposed. In the revised GMDH-type neural networks, the optimum network architecture is automatically organized so as to minimize the prediction error criterion defined as PSS (prediction sum of squares) by using the heuristic self-organization method. On the other hand, in the case of the conventional multi-layered neural networks, the prediction error criteria defined as PSS and AIC (Akaike's information criterion) can not be used to determine the optimum network architecture. In the revised GMDH-type neural networks proposed in this paper, the structural parameters such as the number of the neurons, the useful input variables and the number of the feedback loop calculations are automatically determined so as to minimize PSS. Furthermore, the revised GMDH-type neural networks can identify the radial basis function networks accurately because the complexity of the neural networks is increased gradually by the feedback loop calculations.