Modeling of fuzzy-neural systems using the coevolutionary algorithm

This paper presents a novel hybridization of the fuzzy logic, the neural network and the coevolutionary algorithm for building a fuzzy-neural system (or a Mamdani fuzzy system) from data. The novel hybridization uses the coevolution of many species, and proposes the coevolution of groups of similar species, both for the optimization of the structure of the fuzzy-neural network. In the fuzzy-neural network the coevolution changes the number of nodes and their parameters, which indirectly change the number of fuzzy sets and their parameters and the number of rules and their parameters. Specific backpropagation that supports the Mamdani type of fuzzy system is proposed for small size optimization of fuzzy sets parameters. The backpropagation is active while the average absolute error is small, otherwise the backpropagation stops and the coevolution is active. To be able to guide the coevolution based on three criteria the coevolution uses three level of the fitness. It is possible to control the overfitting through these criteria. The proposed hybridization and its Matlab implementation can be used for creating Mamdani fuzzy-neural systems or simply Mamdani fuzzy systems, from data. This is an alternative for ANFIS and similar hybridizations. It offers to users possibility of building a Mamdani fuzzy-neuro system from data, automatically, with optimizing the number of rules, controlling the overfitting, working with large data sets and many variables, using simple triangular fuzzy sets, the result with high function approximation and good knowledge presentation, the result that can be used as the Mamdani fuzzy system (Matlab FIS object) or as the neural network (internal presentation format and feedforward function). This paper also presents the results of testing with the Wisconsin Breast Cancer Database, from UCI Machine Learning Repository ([http://archive.ics.uci.edu/ml]. Irvine, CA: University of California, School of Information and Computer Science). Key-Words: fuzzy logic, neural network, coevolutionary algorithm, Mamdani fuzzy system, hybridization