On improving the classification of imbalanced data

Abstract

Mining of imbalanced data is a challenging task due to its complex inherent characteristics. The conventional classifiers such as the nearest neighbor severely bias towards the majority class, as minority class data are under-represented and outnumbered. This paper focuses on building an improved Nearest Neighbor Classifier for a two class imbalanced data. Three oversampling techniques are presented, for generation of artificial instances for the minority class for balancing the distribution among the classes. Experimental results showed that the proposed methods outperformed the conventional classifier.