CONSTRUCTION OF CONVOLUTION CRITERIA FOR REGRESSION MODELS

Abstract

Background. The paper proposes a method for constructing a linear convolution of criteria for the adequacy of regression equations based on information generated during the competition of models. Cases of compatibility and inconsistency of the system of inequalities generated by this information are considered. In the first of them, it is proposed to search for a PC-solution of this system, which maximizes its resolution, and in the second, to determine a quasi-solution of the system that minimizes the indicated inconsistency. Materials and methods. To achieve the goal, an apparatus for solving linear programming problems with mixed constraints is involved. Results. As a result of solving the formed linear programming problems, the coefficients of the linear convolution of the partial criteria for the adequacy of the regression model are estimated based on the individual preferences of the researcher regarding the comparative significance of these criteria. Conclusions. The constructed convolution of criteria is based on the individual preferences of the researcher regarding the comparative significance of these criteria and can be repeatedly used by him when constructing models of other objects.

Keywords

• regression model
• linear convolution
• adequacy criteria
• pc-solution
• quasisolution
• linear programming problem