Influence of sample division on the quality of modeling and forecasting of real processes. Nina Kondrashova. IWIM, Prague, 2007.

 Article (in pdf)

Abstract. Sample division and a criterion for choice of the best division are important elements in GMDH algorithms. Sample divisions effective in the tasks of approximation, extrapolation and forecasting are considered in the paper. The main attention was placed to quasi-optimal sample division which enables to enhance the extrapolation and forecasting precision in combination with an adaptive prognosis. Some set of sample division methods allows to choose a proper technology for every task taking into account the object features.

Keywords. GMDH, sample division, quasi-optimal, division by dispersion, approximation, extrapolation, forecast.

References.