AdaBoost ensemble of DCOG rough-neuro-fuzzy systems

Abstract

Neural networks are able to perfectly fit to data and fuzzy logic systems use interpretable knowledge. These methods cannot handle data with missing or unknown features what can be achieved easily using rough set theory. In the paper we incorporate the rough set theory to ensembles of neuro-fuzzy systems to achieve better classification accuracy. The ensemble is created by the AdaBoost metalearning algorithm. Our approach results in accurate classification systems which can work when the number of available features is changing. Moreover, our rough-neuro-fuzzy systems use knowledge comprised in the form of fuzzy rules to perform classification. Simulations showed very clearly the accuracy of the system and the ability to work when the number of available features decreases. © 2011 Springer-Verlag Berlin Heidelberg.