Targeting class imbalance problem using GAN
Hitesh Bhagwani, Sonali Agarwal, Ashwini Kodipalli, Roshan Joy Martis
Abstract
Imbalanced data categorization is inescapable, and it has an impact on the model’s classification problem, which can cause false results. The purpose of this paper is to propose a generative adversarial network (GAN) for restoring balance in imbalanced datasets. This is a challenge since the limited minority data may not be sufficient for GAN training. The proposed article overcomes this issue by adversarial training all available data of minority and majority classes. The data for the minority class is generated using the generative model, which learns all of the useful features from the majority class. The generator in the GAN generates realistic-looking minority class samples. To validate the given method’s classification performance, experiments are performed on a credit card fraud detection dataset. This paper uses a Generative Adversarial Network to give an appropriate solution for imbalanced data classification.