INVESTIGATION OF THE STABILITY OF METHODS OF SELECTING THE OPTIMAL MODEL OF MULTIPLE LINEAR REGRESSION IN THE CASE WHEN INDEPENDENT VARIABLES ARE OBSERVED WITH ERRORS

Abstract

The situation is considered when the input variables of the regression model contain errors. For such a case, the stability of the various criteria commonly used to select the regression model optimal from the point of view of the set of input variables is carried out. The study was carried out by simulation using a software specially developed in the MATLAB environment.
It is shown that the most stable to the presence of errors in input variables is an algorithm based on the method of sequential elimination.