PROBLEM OF THE STRENGTHENING OF REGRESSION-ANALYSIS HYPOTHESES

Based on Zagoruiko's approach to the ''empirical prediction'' problem, methods of searching for an informative set of attributes in regressi on analysis have been considered. Various zero regression coefficient hypotheses and the corresponding submodels are described. The first hy pothesis is referred to as stronger than the second one, if either of them is equivalent to the complete model and the number of zero coeffi cients is larger in the former hypothesis. A hypothesis-strengthening algorithm is an algorithm that enables one, for certain sets of backgr ound data at least, to define a hypothesis stronger than the original complete model. The feasible regression method, branch-and-bound metho d, step-by-step regression, random adaptation search and others can be represented in the form of a hypothesis-strengthening algorithm. The authors give an overview of their previously published results in inac cessible publications.