Evolving Fuzzy Systems from Data: the Computational eFSLab(oratory)
Authors
Abstract
A software laboratory is presented to supportdata-driven approaches for fuzzy rules derivation, reducing
redundancy and lowering the complexity in the obtained
membership functions, giving some semantic meaning to the
fuzzy system. Implemented on-line techniques for merging
membership functions and techniques for rule base
simplification, improve interpretability of the produced fuzzy
models, contributing to the transparency of the obtained rules.
The application, developed in Matlab environment, and public
under GNU license, is demonstrated in the well known
benchmark - the Box-Jenkins time series prediction.