Improving the performance of the k rare class nearest neighbor classifier by the ranking of point patterns

Abstract

In most real life applications of classification, samples are imbalanced. Usually, this is due to the difficulty of data collection. Large margin, or instance based classifiers suffer a lot from sparsity of samples close to the dichotomy. In this work, we propose an improvement to a recent technique developed for rare class classification. The experimental results show a definite performance gain.