GMDH-based forecasting for the estimation of medicines efficiency. Andriy Pavlov, Nina Kondrashova

Abstract. Influence of different initial set division methods on GMDH forecasting models accuracy is researched. Investigations are carried out within an application task in medical diagnostic sphere, namely in the task of medicines efficiency estimation. Three forecasting models construction schemes and four different set division types are investigated. Models are obtained by the two-stage division modeling algorithm and its modifications proposed. Several studies directions in forecasting models construction are analyzed. Criteria and results of numerical experiments are given. Analysis result showed that best forecasting models pair were obtained as sum of trend model and model of remains, utilizing quasi-optimal set division of the incomplete subset space with adaptive forecast scheme. The models pair satisfied the specified ±5,1% error range of the patients main group set with 0,04 significance level.

Keywords. Forecasting model, GMDH algorithm, set division, adaptive forecast, thiol-disulfide ratio (TDR).

References.

1. Nina Kondrashova, Andriy Pavlov.: GMDH-based Forecasting of the Test Results of Blood Samples in Task of the Effective Medicines Estimation. Proceedings of 2nd international conference on Inductive modeling, Sept.15-19, Kyiv, p. 295-300, 2008, ISBN 978-966-02-4889.

2. Kondrashova N.V., Pavlov A. V.: Forecasting models for the estimation of thiol-disulfide ratio by the three-stage sample division. Control Systems and Computers, 2008, №3. (in Russian).

3. Stepashko V.S. The structural identification of forecasting models in conditions of the design of experiments.– Automatica, 1992, №1, p. 26-35. (in Russian).

4. Kondrashove N.V. Forecasting models accuracy comparison when vary dividing set. Coll. Modeling, 2008. – C.128-136. (in Russian)

5. Nina Kondrashova, Andriy Pavlov, Yaroslav Pavlov.: Optimization of Forecasting Models for Testing Blood Samples by Estimation of Thiol-Disulfide Diagrams. IWIM 2007 Workshop proceedings, 160-165, 2007