Cascade neo-fuzzy neural network in the forecasting problem at stock exchange
DOI:
https://doi.org/10.20535/SRIT.2308-8893.2017.2.09Keywords:
forecasting, stock exchange, cascade neo-fuzzy neural network, FNN ANFISAbstract
A forecasting problem at the stock exchange is considered. For its solution the application of a cascade neo-fuzzy neural network (CNFNN) is suggested. The architecture of the neo-fuzzy neuron and architecture of CNFNN is presented. Training algorithms of CNFNN in packet mode and on-line are described and discussed. The experimental investigations of CNFNN for market index forecasting at the German stock exchange are carried out. During experiments, the number of cascades, inputs, linguistic terms, and the training-to-test ratio of samples were varied. In the experiments, the optimal values of the aforesaid parameters of the training algorithm were found. The comparative experiments estimating forecasting efficiency of the cascade neo-fuzzy neural network and FNN ANFIS were carried out.References
L'juis K.D. Metody prognozirovanija ekonomicheskih pokazatelej: per. s angl. / K.D. L'juis. – M.: Finansy i statistika, 1986. – 133 s.
Zgurovskij M.Z. Osnovy vychislitel'nogo intellekta / M.Z. Zhurovskyj, Ju.P. Zajchenko. – K.: Nauk. dumka, 2013. – 406 s.
Bodyanskiy Ye. The cascade NFNN learning algorithm / Ye. Bodyanskiy, Ye. Viktorov, I. Pliss // Visn. Uzhhorod. nats. un-tu. Serija " Matematyka i informatyka". – 2008. – Vyp. 17. – S. 48–58.
Bodyanskiy Ye. Neo-fuzzy neural network structure optimization using GMDH for solving forecasting and classification problems / Ye. Bodyanskiy, Yu. Zaychenko, E. Pavlikovskaya etc // Proc. Int. Workshop on Inductive Modeling 2009. Krynica, Poland, 2009. – P. 77–89.