Learning hybrid neuro-fuzzy classi(er models from data: to combine or not to combine?

To combine or not to combine? This very important question is examined in this paper in the context of a hybrid neuro-fuzzy pattern classi(er design process. A general fuzzy min–max neural network with its basic learning procedure is used within (ve di3erent algorithm-independent learning schemes. Various versions of cross-validation and resampling techniques, leading to generation of a single classi(er or a multiple classi(er system, are scrutinised and compared. The classi(cation performance on unseen data, commonly used as a criterion for comparing di3erent competing designs, is augmented by further four criteria attempting to capture various additional characteristics of classi(er generation schemes. These include: the ability to estimate the true classi(cation error rate, the classi(er transparency, the computational complexity of the learning scheme and the potential for adaptation to changing environments and new classes of data. One of the main questions examined is whether and when to use a single classi(er or a combination of a number of component classi(ers within a multiple classi(er system. c © 2003 Elsevier B.V. All rights reserved.