Parallel GMDH algorithm with successive selection of informative arguments for effective solving high-dimensional modelling problems. Oleksandr Samoilenko, Serhiy Yefimenko, Volodymyr Stepashko

Abstract. In recent investigations we considered GMDH algorithms for solving problems with a large number of arguments based on successive selection of the most informative arguments. These algorithms build models very quickly but the accuracy of obtained models is often not very high. The models quality being built in such a way depends on the quality of informative arguments selection. To increase this quality in the algorithm with successive selection, the inverted structures are used. As a result we have the quality enhancement but the speed of the model building is somewhat reduced. To solve this problem the parallel algorithm is implemented. Thus the main goals of our work are improvement of the method with successive selection of arguments and the parallel algorithm implementing in this method for the enhancement of the quality and effectiveness of the informative arguments definition. This paper considers the main aspects of the algorithm and results of its performing. The test experiments for the parallel implementation of algorithm with successive selection of arguments using inverted structures on a cluster system are carried out. The results of these experiments confirm effectiveness of the method.

Keywords. Inductive modelling, combinatorial GMDH algorithm, parallel computing, cluster system, successive selection, informative arguments.

References.

1. Ivakhnenko A.G.: Ivakhnenko G.A., Savchenko E.A., and Wunsch D.: Problems of Further Development of GMDH Algorithms: Part 2. Pattern Recognition and Image Analysis , Vol. 12, № 1, 2002, p.6-18.

2. Ivahnenko A.G., Ivahnenko G.A., Savchenko E.A.: Conception of the successive algorithmic approaching (lowering) to the exact decision of interpolation tasks of artificial intelligence. Cybernetics and computing engineering, № 127, 2000, p.47-58. (In Russian)

3. Stepashko V.S., Koppa Y.V.: The Experience of application of the ASTRID system for the design of economic processes from statistical data. Cybernetics and computing engineering. - 1998. - V.117. – p. 24-31. (In Russian)

4.  https://icybcluster.org.ua/