Multiple Linear Regression Based on Coefficients Identification Using Non-iterative SGTM Neural-like Structure
- Author(s)
- Ivan Izonin, Roman Tkachenko, Natalia Kryvinska, Pavlo Tkachenko, Michal Greguš
- Abstract
In the paper, a new method for solving the multiple linear regression task via a linear polynomial as a constructive formula is proposed. It is based on the use of high-speed SGTM Neural-Like Structure. This linear non-iterative computational intelligence tool is used for identification of polynomial coefficients. As a result of the implementation of the learning algorithm and applied the matrix of test signals to the trained SGTM, the identification of the linear polynomial coefficients is carried out. A further solution of the task occurs by searching a dependent variable using the obtained polynomial. The results of the method have been tested on the task of the output of the electric power prediction of the combined-type factory. The method ensures the identification of five polynomial’s coefficients at the high speed, which ensures high accuracy of the solution. Based on the comparison with known regression analysis methods, the highest accuracy of the work has been established. The transition from neural-like structure to the solution of the task in the form of a linear polynomial provides the possibility for the simple interpretation of the result of the regression or classification tasks. That does not require high qualifications from the user. In addition, the developed method, based on the repetition of training outcomes and the lack of debugging and parameter selection procedures, allows synthesizing linear polynomial for complex models that use various non-linear extensions of SGTM inputs while preserving the accuracy of their operation. The proposed approach can be used in the fields of medicine, economics, materials science, service sciences etc., for fast and accurate solution of regression or classification tasks with the possibility of easy interpretation of the result.
- Organisation(s)
- Department of Marketing and International Business
- External organisation(s)
- Lviv Polytechnic National University, Comenius University Bratislava, IT STEP University
- Pages
- 467-479
- No. of pages
- 13
- DOI
- https://doi.org/10.1007/978-3-030-20521-8_39
- Publication date
- 2019
- Peer reviewed
- Yes
- Austrian Fields of Science 2012
- 502052 Business administration, 502050 Business informatics
- Keywords
- ASJC Scopus subject areas
- Theoretical Computer Science, General Computer Science
- Portal url
- https://ucrisportal.univie.ac.at/en/publications/87032302-ce2d-4b66-9464-a6eddc67e98c