An Extended-Input GRNN and its Application

Autor(en)

Ivan Izonin, Natalia Kryvinska, Roman Tkachenko, Khrystyna Zub, Pavlo Vitynskyi

Abstrakt

A new extended-input General Regression Neural Network scheme is proposed. The main objective of such a step was to increase the accuracy of the regression tasks. Such an extension is based on using of the Ito decomposition. This scheme is more appropriate in comparison with existing ones and provides an increase of the prediction accuracy due to the high approximation properties of this decomposition. The developed ANN is used to solve the missing data recovery task. This real dataset was collected by the IoT device, and it is characterized by a large number of passes. A number of practical experiments were carried out on setting of the optimal parameters of the proposed scheme. It has been established that the values ​​of Gaussian functions deviations greater than 0.1 greatly increase the errors of extended-input GRNN. In addition, the larger than the second order Ito decomposition does not improve the accuracy of the ANN and substantially increases the duration of its use. Experimentally established the highest accuracy of the developed ANN in comparison with existing methods.

Organisation(en)

Institut für Marketing und International Business

Externe Organisation(en)

Lviv Polytechnic National University

Journal

Procedia Computer Science

Band

160

Seiten

578-583

ISSN

1877-0509

DOI

https://doi.org/10.1016/j.procs.2019.11.044

Publikationsdatum

2019

Peer-reviewed

Ja

ÖFOS 2012

502050 Wirtschaftsinformatik

Schlagwörter

ASJC Scopus Sachgebiete

Allgemeine Computerwissenschaft

Link zum Portal

https://ucrisportal.univie.ac.at/de/publications/d6178c48-1fe4-4f67-bfd9-ee1512179f3d