COMBINING ALTERNATIVE OPTIONS OF THE REGRESSION MODEL BASED ON THE BEHAVIOR CONSISTENCY CRITERION

Abstract

The purpose of the study is to develop an algorithm for calculating the coefficients of a convex combination of alternative variants of a regression model of a complex object, based on the use of the criterion of behavior consistency between the actual and calculated values of the output variable, specified in continuous form, introduced in previous works by one of the authors. Methods. To solve the problem formulated in the paper, the authors use both the traditional criteria for model adequacy in regression analysis – multiple determination, Fisher, average relative error of approximation – and the behavior consistency criterion previously developed by one of the authors. Results. The study demonstrates the application of the developed method to create an ensemble of regression models for the construction of a mathematical model of the gross domestic product of the Russian Federation. This approach, due to its invariance to the nature of the analyzed systems, does not require special adaptation in the study of objects of technical nature. Conclusions. The proposed algorithm for combining alternative options for a regression model of an object, based on the use of a behavior consistency criterion between the actual and calculated values of the output variable, can be effectively used in the study of complex systems of various natures.