Evolving Hybrid GMDH-Neuro-Fuzzy Network and its Applications

In the paper the evolving GMDH-neuro-fuzzy systems (Group Method of Data Handling) are presented. The main advantage of these systems is small number of tuning parameters. It simplifies the training algorithms and decrease training time comparing with classical evolving fuzzy neural networks (EFNN). The architecture and training algorithms for this system are considered. The experimental investigations of the proposed GMDH-neuro-fuzzy systems were carried out in the problem of stock prices forecasting and comparisons with GMDH algorithm and the cascade neuro-fuzzy network were performed that enable their efficiency estimation.