Title: Robust extraction of fuzzy rules with artificial neural network based on fuzzy inference system

Authors: Robert Czabanski; Michal Jezewski; Janusz Jezewski; Janusz Wrobel; Krzysztof Horoba

Addresses: Institute of Electronics, Silesian University of Technology, ul. Akademicka 16, 44-101 Gliwice, Poland. ' Institute of Electronics, Silesian University of Technology, ul. Akademicka 16, 44-101 Gliwice, Poland. ' Department of Signal Processing, Institute of Medical Technology and Equipment, ul. Roosevelta 118, 41-800 Zabrze, Poland. ' Department of Signal Processing, Institute of Medical Technology and Equipment, ul. Roosevelta 118, 41-800 Zabrze, Poland. ' Department of Signal Processing, Institute of Medical Technology and Equipment, ul. Roosevelta 118, 41-800 Zabrze, Poland

Abstract: The paper presents a method of parameters estimation for artificial neural network based on fuzzy inference system (ANNBFIS). It is based on deterministic annealing, ε-insensitive learning by solving a system of linear inequalities and robust fuzzy c-means clustering. The proposed algorithm allows to improve the neuro-fuzzy modelling quality by increasing the generalisation ability and outliers robustness. To find the unknown number of fuzzy rules we proposed the procedure of robust clusters merging. The performance of the learning method is demonstrated through the benchmark sunspot prediction problem.

Keywords: fuzzy rules extraction; signal prediction; neuro-fuzzy systems; robust methods; artificial neural networks; ANNs; fuzzy inference systems; linear inequalities; fuzzy c-means clustering; modelling; sunspot prediction.

DOI: 10.1504/IJIIDS.2012.045115

International Journal of Intelligent Information and Database Systems, 2012 Vol.6 No.1, pp.77 - 92

Received: 30 Jul 2010
Accepted: 08 Dec 2010

Published online: 16 Aug 2014 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article