The Recursive Fuzzy Hypercube

Abstract

A method for automated fuzzy modeling of classification problems by means of supervised learning, is discussed. The method is based on a computational architecture called a recursive fuzzy hypercube. The method consists of several steps including: the determination of a good fuzzy set design based on a 'local entropy' analysis of the various features, development of the fuzzy rule base and identification of conflict cells, and the treatment of each conflict cell as a sub-problem, reapplying the method from the beginning. The ability to apply the method recursively to smaller sub-problems leads to very good classification accuracies.