Using Clustering for Generating Diversity in Classifier Ensemble

In the past decade many new methods were proposed for creating diverse classifiers due to combination. In this paper a new method for constructing an ensemble is proposed which uses clustering technique to generate perturbation in training datasets. Main presumption of this method is that the clustering algorithm used can find the natural groups of data in feature space. During testing, the classifiers whose votes are considered as being reliable are combined using majority voting. This method of combination outperforms the ensemble of all classifiers considerably on several real and artificial datasets.