Investigating Noise Tolerance in Generalised Nearest Neighbour Learning
Alexander U. J. Wong
Department of Computer Science
University of Canterbury
Abstract
In this report we investigate the effects of integrating techniques and methods that tolerate noise well in nearest neighbour systems into generalised nearest neighbour systems and find whether or not this similarly helps in their tolerance of noise. We use Nearest Neighbour with Generalised Exemplars (NNGE) as our base generalised nearest neighbour system and create alternative variations, k-NNGE and NNGE-S, which we predict will perform better than the original NNGE in noisy domains. Our findings show that this is not in fact the case but insightful discoveries from this outcome has resulted in a beneficial investigation.
