Investigation of selective properties of the error bias criterion. Liudmyla Somina,Ievgeniia Savchenko, Volodymyr Stepashko

Abstract. Selective properties of a new error bias criterion are investigated first analytically. It is shown that the same optimal model can be chosen by the new criterion as by classical solution bias criterion. This criterion is more sensitive to change of model complexity then classical. The numerical analysis of the error bias criterion for different levels of noise in data is carried out. The example shows that error bias criterion has the main property of effective external criterion: property of noise-immunity. It is shown that sequential using of the regularity criterion firstly and the error bias criterion secondly are the most reasonable for selection of an optimal complexity model.

Keywords. Inductive modelling, GMDH, regularity criterion, noise-immunity, error bias criterion, Pareto region.

References.

1. Ivakhnenko A.G., Ivakhnenko G.A., Savchenko E.A. GMDH Algorithm for Optimal Model Choice by the External Error Criterion with the Extension of Definition by Model Bias and its Applications to the Committees and Neural Networks // Pattern Recognition and Image Analysis. 2002. V.12, N 4. P.347-353.

2. Ivakhnenko A.G., Stepashko V.S. Noise-immunity of modelling: Kiev.- Naukova Dumka. 1985. 216 p. (in Russian).

3. Ivakhnenko A.G., Savchenko E.A. Investigation of Efficiency of Additional Determination Method of the Model Selection in the Modeling Problems by Application of the GMDH Algorithm // Journal of Automation and Information sciences. Begell House., Inc. Publishers. 40, N 3, 2008. pp.47-58.

4. Savchenko I. Double-criterion Choice of an Optimal Model in GMDH Algorithms // Proceedings of 2nd International Conference on Inductive Modelling, September 15-19, 2008, Kyiv, p. 23-27.

5. Stepashko, V.S. Method of Critical Variances as Analytical Tool of Theory of Inductive Modeling // Journal of Automation and Information Sciences, 2008, 40, No. 3, pp. 4-22.