Fuzzy systems modeling in practice

Instead of describing a fuzzy modeling algorithm that is new, powerful, robust, and with outstanding learning abilities, the objective of this paper is to point out four important topics usually “ignored” in fuzzy model’s design. These are: 1. The generalization ability of the fuzzy model. 2. The appearance of empty rules at a fuzzy model whose conclusions could not be extracted. 3. The presence of noise as source of ambiguity to the fuzzy model. 4. The in4uence of training set size on learning performance. These topics are analyzed and discussed by modeling a linear functional relation using a basic learning algorithm. These conditions allow a better understanding, visualization, and separation of the causes a5ecting the fuzzy models performance when using this learning algorithm. Results show that it is important to understand what information can be obtained from a previous analysis of the training data which can help to design reasonable and e7cient fuzzy models to work in practical environments. c © 2001 Elsevier Science B.V. All rights reserved.