OPTIMIZATION OF STRUCTURE OF LINEAR-REGRESSION PREDICTION MODELS UNDER CONDITION OF UNCERTAINTY

Solution of the problem on a minimax reduction of the structure of the linear regression predicting models under stochastic and nonstochastic uncertainty is considered. It is shown that solution of the initial problem for the quadratic functions of losses and a class of linear estimates of the model parameters is reduced to the check-up of certain inequalities whose, quantity coincides with the dimensionality of the vector of parameters if there is an apriori information relative to the unkown characteristics in the form of their reference to the preset closed restricted sets.