A neurofuzzy selfmade network with output dependable on a single parameter

Abstract

In some fuzzy systems the number of rules and the membership functions are estimated by designers, often being a tedious task. In this paper we describe a neurofuzzy system (SIMAP) able to build its structure and membership functions using only the input-output data. The system compresses the input-output data, minimizing predictive error by the increment of an input vigilance-parameter, in a similar way to the fuzzy-artmap neural network (G A. Carpenter et al., 1992). In the SIMAP network the output-clusters are weighted to obtain the final output vector, implementing a continuous map. A method for calculating the membership functions in neurofuzzy systems is proposed. These membership functions are used to operate the SIMAP network. The softness of the inference mechanism can be controlled adjusting a single fuzziness-parameter p.