Enhanced MIA-GMDH Algorithm. Petr Buryan. IWIM, Prague, 2007.
Download in .pdf
Abstract. The paper presents a new methodology which is an enhancement of MIA algorithm of self-organizing polynomial Group Method of Data Handling (GMDH). Classical MIA algorithm suffers mainly by quick loss of the layer diversity resulting in almost homogenous layer output which strongly prohibits reaching any improvement in the next layers of the network and has negative impacts on the stability of transfer functions of the nodes (gained by least mean squares method). The impact is mainly non-stable behaviour and loss in quality of the output of the model as a whole. Several specific improved features were therefore applied in order to improve the behaviour of the algorithm. The enhancements described bellow are mainly semi-randomized selection approach to layer pruning, coefficient rounding and thresholding schemes. The usefulness of proposed enhancements is supported by experimental results of time series analyses.
Keywords. Inductive modelling, GMDH algorithm, time series prediction, self-organizing polynomial networks.
References.
- Ivakhnenko A. G.: Polynomial theory of complex systems. IEEE Transactions on Systems, Man, and Cybernetics, SMC-1(1):364378, 1971.
- H. R. Madala, A. G. Ivakhnenko, Inductive Learning Algorithms for Complex Systems Modeling, CRC Press, Boca Raton, 1994
- P. Buryan, Time Series Analysis by Means of Enhanced GMDH Algorithm, Dissertation Thesis, CTU Prague, 2006.
- Taylor, S., Modeling Financial Time Series, Wiley, New York, NY, 1986
- Arlt, J. Finanční časové řady, Grada,Praha 2003
- Rosenberg, M.: Currency Forecasting: A Guide to Fundamental and Technical Models of Exchange Rate Determination, Irwin Professional Publishing, 1996
- Kröse, B., Smagt, P. van der, An introduction to neural networks. University of Amsterdam, November 1996
- Mackey, M. C., and L. Glass, Oscillation and chaos in physiological control systems, Science, vol. 197, pp. 287-289, 1977.
- PlanetMath? [online] http://planetmath.org/encyclopedia/MatrixConditionNumber.html, latest revision 23-05-2007
- Kordík P. Group of Adaptive Models Evolution, Technical Report, ČVUT Praha 2005. [online] http://neuron.felk.cvut.cz/game/index.html, latest revision 23-05-2007
- Maintaining Diversity in Population of Evolved Models, Drchal J., norek M., Kordík P. In Proceedings of 40th Spring International Conference MOSIS 06, Modelling and Simulation of Systems. Ostrava: MARQ, 2006, vol. 1, p. 113-120. ISBN 80-86840-21-2.
