Improving k Nearest Neighbor with Exemplar Generalization for Imbalanced Classification

A k nearest neighbor (kNN) classi er classi es a query in- stance to the most frequent class of its k nearest neighbors in the training instance space. For imbalanced class distribution, a query instance is of- ten overwhelmed by majority class instances in its neighborhood and likely to be classi ed to the majority class. We propose to identify exem- plar minority class training instances and generalize them to Gaussian balls as concepts for the minority class. Our k Exemplar-based Nearest Neighbor (kENN) classi er is therefore more sensitive to the minority class. Extensive experiments show that kENN signi cantly improves the performance of kNN and also outperforms popular re-sampling and cost- sensitive learning strategies for imbalanced classi cation.