Classification by fuzzy integral: Performance and tests

An introductory explanation about the approach is given, a lower bound of the minimal number of training samples is found, and it is shown that a minimum squared error criterion leads to the best approximate for the optimal Bayes classifier.

Abstract

This paper presents an attempt to characterize the performance of methods of classification based on fuzzy integral. After an introductory explanation about the approach, a lower bound of the minimal number of training samples is found, and it is shown that a minimum squared error criterion leads to the best approximate for the optimal Bayes classifier. Some tests on simulated and real data are provided.