ProWSyn: Proximity Weighted Synthetic Oversampling Technique for Imbalanced Data Set Learning

摘要

An imbalanced data set creates severe problems for the classifier as number of samples of one class (majority) is much higher than the other class (minority). Synthetic oversampling methods address this problem by generating new synthetic minority class samples. To distribute the synthetic samples effectively, recent approaches create weight values for original minority samples based on their importance and distribute synthetic samples according to weight values. However, most of the existing algorithms create inappropriate weights and in many cases, they cannot generate the required weight values for the minority samples. This results in a poor distribution of generated synthetic samples. In this respect, this paper presents a new synthetic oversampling algorithm, Proximity Weighted Synthetic Oversampling Technique (ProWSyn). Our proposed algorithm generate effective weight values for the minority data samples based on sample's proximity information, i.e., distance from boundary which results in a proper distribution of generated synthetic samples across the minority data set. Simulation results on some real world datasets shows the effectiveness of the proposed method showing improvements in various assessment metrics such as AUC, F-measure, and G-mean.