Synthetic Data Generation for Imbalanced Datasets Using GAN Variants
Keywords:
Data shortage, Imbalanced datasets, Generative Adversarial Networks, Wasserstein GAN, Conditional GAN, Data augmentation, Machine learning, Classification performance..Abstract
This paper examines GANs and their derivatives, WGAN and CGAN, for the synthetic generation of data – which, to the best knowledge of the author, is probably the best remedy to the imbalance That can positively boost the value of minority classes that are often affected by this imbalance and consequently enhance a given machine learning models trained on imbalanced data, to generalize well in terms of accuracy and performances. The current study gives some of the key advantages, limitations, and opportunities required using GANs while testing the models in various simulated and real-world applications. This indicates that it is indeed possible to attain much further enhanced classifier performance when using other more enhanced types of GAN.
