Solving an Optimization Problem for Estimating Fully connected Linear Regression Models

34

Abstract

This article is devoted to the problem of estimating fully connected linear regression models using the maximum likelihood method. Previously, a special numerical method was developed for this purpose, based on solving a nonlinear system using the method of simple iterations. At the same time, the issues of choosing initial approximations and fulfilling sufficient conditions for convergence were not studied. This article proposes a new method for solving the optimization problem of estimating fully connected regressions, similar to the method of estimating orthogonal regressions. It has been proven that, with equal error variances of interconnected variables, estimates of b-parameters of fully connected regression are equal to the components of the eigenvector corresponding to the smallest eigenvalue of the inverse covariance matrix. And if the ratios of the error variances of the variables are equal to the ratios of the variances of the variables, then the b-parameter estimates are equal to the components of the eigenvector corresponding to the smallest eigenvalue of the inverse correlation matrix, multiplied by the specific ratios of the standard deviations of the variables. A numerical experiment was carried out to confirm the correctness of the developed mathematical apparatus. The proposed method for solving the optimization problem of estimating fully connected regressions can be effectively used when solving problems of constructing multiple fully connected linear regressions.

General Information

Keywords: fully connected linear regression model, maximum likelihood estimation, numerical method, optimization, orthogonal regression, correlation matrix, eigenvector, eigenvalue

Journal rubric: Optimization Methods

Article type: scientific article

DOI: https://doi.org/10.17759/mda.2024140108

Received: 15.01.2024

Accepted:

For citation: Bazilevskiy M.P. Solving an Optimization Problem for Estimating Fully connected Linear Regression Models. Modelirovanie i analiz dannikh = Modelling and Data Analysis, 2024. Vol. 14, no. 1, pp. 121–134. DOI: 10.17759/mda.2024140108. (In Russ., аbstr. in Engl.)

Information About the Authors

Mikhail P. Bazilevskiy, PhD in Engineering, Associate Professor, Department of Mathematics, Irkutsk State Transport University (ISTU), Irkutsk, Russia, ORCID: https://orcid.org/0000-0002-3253-5697, e-mail: mik2178@yandex.ru

Metrics

Views

Total: 57
Previous month: 7
Current month: 0

Downloads

Total: 34
Previous month: 3
Current month: 0