Learning membership functions from examples

Describes the adaptation of a crisp induction algorithm, called constrained generalization, to the building of fuzzy rules. This approach bridges the gap between machine learning and fuzzy logic. An application is learning membership functions from examples, as well as estimating a real-valued attribute. By means of membership functions, measurable information is possibly correctly translated within linguistic qualifiers. The acceptance of linguistic qualifiers is generally context-dependent. Therefore, a method for automatically designing membership functions from examples is presented. This method, inspired by supervised learning, also applies when the target concept depends on several attributes. It then builds fuzzy rules. The approach is validated on a real-world problem, predicting the elastic limit of new materials from a database about trials on composite materials