|Title of the article||
ENTROPY-NEURAL NETWORK METHOD FOR ELIMINATING INCONSISTENCIES IN EXPERT ASSESSMENTS
Igor A. Kubasov, Doctor of technical sciences, associate professor, professor of the sub-department of information technologies, Academy of Management of the Ministry of Internal Affairs of Russia, (8 Zoya and Alexander Kosmodemyanskikh street, Moscow, Russia), E-mail: email@example.com
Background. The paper investigates ways to solve the problem of eliminating the possible inconsistency of opinions of different experts when evaluating the rating of complex systems compared in terms of reliability and quality parameters. This problem arises due to the fact that the choice of an integral indicator and the establishment of an integral criterion for evaluating complex systems that take into account the values of reliability and quality parameters may be different for different experts. At the same time, a situation is theoretically possible when the votes of experts "for" and "against" are divided exactly in half and then the decision to form a rating of complex systems will be unstable. Materials and methods. The expediency of comparing several leaders of the rating of the compared complex systems is justified, not among themselves, but with the average system of the considered set of complex systems. At the same time, the resulting effect of blurring the difference between the leaders of the rating is eliminated by training a neural network (with a large number of neurons) to separate the compared leaders, and then controlling the Hamming distances and/or the entropy difference between the leaders in relation to the distances to the average system. Results. The application of the proposed entropy-neural network method allows us to objectively determine the first leader of the rating by the maximum jump in the entropy of code responses to the second leader in order. Conclusions. A new method is proposed to eliminate the possible inconsistency of opinions of different experts, which allows to obtain an objective result on the formation of a rating of complex systems, based on fully automatic training of neural networks and automatic classification by neural networks of all compared complex systems.
complex systems, expert evaluation, artificial neurons, Hamming distance, entropy of code states
Дата обновления: 16.02.2022 13:43