A Positive-biased Nearest Neighbour Algorithm for Imbalanced Classification

摘要

The k nearest neighbour (kNN) algorithm classifies a query instance to the most frequent class among its k nearest neighbours in the training instance space. For imbalanced class distribution where positive training instances are rare, a query instance is often overwhelmed by negative instances in its neighbourhood and likely to be classified to the negative majority class. In this paper we propose a Positive-biased Nearest Neighbour (PNN) algorithm, where the local neighbourhood of query instances is dynamically formed and classification decision is carefully adjusted based on class distribution in the local neighbourhood. Extensive experiments on real-world imbalanced datasets show that PNN has good performance for imbalanced classification. PNN often outperforms recent kNN-based imbalanced classification algorithms while significantly reducing their extra computation cost.