Fuzzy If-Then Rules Extraction by Means of ε-Insensitive Learning Techniques Integrated with Deterministic Annealing Optimization Method

This paper introduces the research on possibility of global optimization elements and ε-insensitive learning techniques integration in aim of fuzzy if-then rules extraction quality increase. The new learning algorithm of neuro-fuzzy system with parameterized consequents is introduced. It consists in integration of deterministic annealing and ε iterative quadratic programming method. The proposed algorithm indicates generalization ability and outliers robustness improvement comparing with zero-tolerance learning procedures. To show usefulness of introduced method two numerical experiments concerning system identification and chaotic time series prediction problems are provided. Copyright c © 2005 Yang’s Scientific Research Institute, LLC. All rights reserved.