A genetic-based neuro-fuzzy generator: NEFGEN

Abstract

This paper is concerned with the design of a new generation of intelligent systems. These systems or machines are intelligent if they are able to improve their performance or maintain an acceptable level of performance in the presence of uncertainty. The ability of these systems to examine and modify their behaviors in a limited sense is usually achieved by using techniques such as knowledge-based systems (KBS), artificial neural networks (NNs), fuzzy systems (FSs), and genetic algorithms (GAs). We propose a novel technique called NEFGEN or Neuro-Fuzzy Generator. This hybrid neuro-fuzzy generator is based on the knowledge oriented design (KOD) concept, cooperative neuro-fuzzy systems and genetic algorithms. A classification through a competitive neural network of data examples of the application to be performed provides efficient inference rules as well as adequate fuzzy partitions of the input/output variable domains. The resulting fuzzy system is then optimized using random techniques and genetic algorithm techniques. NEFGEN is proved to be very efficient in designing powerful fuzzy expert systems (FESs) especially in classification and approximation. It is also shown that NEFGEN performance exceeds that of known hybrid neuro-fuzzy systems such as ANFIS, NEFPROX, and NEFCLASS.