Nyzzuf 1

Abstract

Fuzzy-neural network attract an attention by their power of neural networks adaptation and expressive capability of fuzzy rules. Some approaches encode the rules into the neural network, which is then used to fine-tune the parameters [ANFIS]. There are also neural networks designed to work like a fuzzy rules system with the ability to automatically discover the rules [FINN]. However, they are not designed with the rule extraction as the main purpose in mind. Is there a way to modify the current models to make the rule extraction easier? How could the resulting rules be made more readable for a human expert? Do we have to sacrifice the performance to obtain these features? This articles describes a new modification of the current Takagi-Sugeno fuzzy-neural network models to the more human understandable Mamdami linguistic form, introduces a novel double-clustering and shows the performance of the new model on the breast-cancer dataset available from the UCI repository.