The Usage of the k-Nearest Neighbour Classifier with Classifier Ensemble

The objective of this paper is to try and determine the usability of the k-Nearest Neighbour classifier as a base classifier for an ensemble. To do this, five different ensembles are tested on a group of ten varied datasets. The most popular ensembles are taken into consideration, including Bagging, AdaBoost and Random Subspaces, as well as recently introduced algorithm called Feating. Moreover, a new algorithm, proposed by authors of this paper, called Rotation Ensemble, is introduced and its performance is evaluated. The accuracy gain over a single k-Nearest Neighbour classifier as well as in comparison with other ensemble methods displayed by the Rotation Ensemble algorithm seems to be very promising.