Learning Fuzzy Rules from Data

Fuzzy models have been designed to represent approximate or imprecise relationships in complex systems and have been successfully employed in control systems, expert systems, and decision analysis. Classically, fuzzy models were built from human expertise and knowledge of the system being modeled. As systems have grown more complex it has become increasingly difficult to construct models directly from domain knowledge of the system. Recently, learning algorithms have been investigated to construct fuzzy models by developing fuzzy rules through analysis of training data. This research presents a hierarchical architecture for fuzzy modeling and inference that learns the underlying fuzzy rules from training data. The effects of increased granularity and dimensionality on the performance of the system are illustrated and discussed, along with techniques to minimize the adverse impacts of this increased complexity. A short discussion of ongoing research aimed at allowing the combination of human expertise with fuzzy learning strategies is also included.