Fuzzy pattern recognition by fuzzy integrals and fuzzy rules
Résumé
We give an overview of the application of fuzzy rules and fuzzy integrals in classification, presenting the general methodology and illustrating it by giving real applications. Fuzzy rules have been most of the time devoted to fuzzy control, using the so-called “Mamdani rules”. Here we present a broader view of the topic in the framework of possibility theory. It is seen that uncertainty rules are the best-suited ones for modeling human knowledge in pattern recognition. On the other hand, fuzzy integrals, by assigning weights to groups of attributes, permit to define non linear classifiers, with powerful performances.