Constructing multi-layer classifier ensembles using the Choquet integral based on overlap and quasi-overlap functions

Ensembles of classifiers have been receiving much attention lately, they consist of a collection of classifiers that process the same information and their output is combined in some manner. The combination method is probably the most important part in a ensemble of classifiers however, many works found in literature focus mostly on the classification step, using simple approaches on the combination step, such as majority voting. In this paper, we propose a new combination method based on a generalization of discrete Choquet integrals, combining it with quasi overlap functions to aggregate the outputs of classifiers in an ensemble. We also tested the proposed combination approach in a simple ensemble against other methods in literature to verify if there was a significant gain in performance.