Scalable, Fuzzy, Multiclassifier System

This paper deals with the design of scalable, fuzzy adaptive, decision making systems constructed from sets of heterogeneous classifiers. We propose a multiclassifier architecture formed by sparsely connected coalitions of classifiers. Coalitions are defined by fuzzy integral operators on small, but not necessarily disjoint, subsets of classifiers. The small size constraint on individual coalitions is intended for both the interpretability of fuzzy measures and low complexity of fuzzy integral operators. In addition, the sparse connection constraint guarantees realizable good independent coalitions amenable to further information fusion stages. Simple rules regarding the number of coalitions, the number of classifiers per coalition and the number of coalitions where each classifier should participate are presented. Experimental results show the feasibility of our proposal.