Multiple Views in Ensembles of Nearest Neighbor Classifiers

Multi-view classification is a machine learning methodology when patterns or objects of interest are represented by a set of different views (sets of features) rather than the union of all views. In this paper, multiple views are employed in ensembles of nearest neighbor classifiers where they demonstrate promising results in classifying a challenging data set of protein folds. In particular, up to 4.68% increase in accuracy can be achieved, compared to the best result in single-view classification, thus rendering ensembles of nearest neighbor classifiers employing multiple views an attractive research direction.