MODIFIED METHOD OF CONSTRUCTING A MULTIVARIATE LINEAR REGRESSION GIVEN BY A REDUNDANT DESCRIPTION

Abstract

A number of scientific works of Prof. O. A. Pavlov and his disciples is devoted to the development of an original method of efficient estimation of coefficients at nonlinear terms of multivariate polynomial regression given by a redundant description under the conditions of an active experiment. The solution of the formulated problem is reduced to the sequential construction of univariate polynomial regressions (finding efficient estimates for the coefficients at nonlinear terms) and solving the corresponding systems of linear nondegenerate equations, the variables of which are the estimates for coefficients at nonlinear terms of the multivariate polynomial regression given by the redundant description. Thus, the problem was reduced to the estimation of the coefficients at linear terms of a multivariate linear regression given by a redundant description in the conditions of an active experiment. We have proposed an original method of its solution that uses a cluster analysis algorithm. The algorithm’s implementation significantly reduces the enumeration of partial descriptions of multivariate linear regression followed by the finding of the residual sum of squares for each of them. This allows using the chi-squared criterion to build a linguistic variable which value gives a qualitative assessment (high reliability, acceptable reliability, low reliability, unreliability) to the obtained result. The analysis of the computational experiments made it possible to modify the proposed method, which significantly increased its efficiency, first of all, of finding a reliable structure of the sought multivariate linear regression given by the redundant description. The method modification, in particular, has reduced the enumeration of partial descriptions and has led to a more efficient use of the general procedure of the least squares method.