Feature augmentation for imbalanced classification with conditional mixture WGANs

Highlights：

• We propose a framework of data augmentation for addressing imbalanced classification without regards to the modality of data.

• Further, we propose a conditional mixture WGANs (cMWGANs), which can stabilize the training process and address the mode collapse of GANs, thus generating label preserved and diverse features.

• In order to test the augmented ability of the method, we evaluate the method on three uneven benchmark datasets. The empirical results show that our model can achieve significant improvements compared to the model without feature augmentation and other typical models.