Nonnegative set functions in multiple classifier fusion
Xizhao Wang, Huimin Feng · 2005
Fuzzy integral is a valid method for combining multiple classifiers. However in the fusion based on fuzzy integral, how to choose an appropriate fuzzy measure is a difficult but important problem. The system's performance is largely dependent of the fuzzy measure. An appropriate fuzzy measure can make the system's performance better than the best individual classifier, while an inappropriate fuzzy measure will result in worse performance than the individual classifiers. This paper investigates the fusion mechanism based on the fuzzy integral for multiple classifiers, and discusses the impact of fuzzy measures or nonnegative set functions on the fusion. The study is useful to obtain an appropriate fuzzy measure for improving the performance of the system.