Fraud detection in the insurance business is a supervised learning classification problem where either a claim is fraudulent or non-fraudulent. A number of models were developed to address this such as random forests, neural networks, and many others. However, the critical issue in building these models is – how to tackle the inherent data imbalance in the given dataset, which is more prominent in the case of insurance frauds. Popular data-imbalance techniques such as SMOTE, MWMOTE, ADASYN, etc. would help address this issue but are not adequate enough to apply in health insurance data. In this work, we are presenting an innovative way of using Generative Adversarial Networks (GANs) to handle the data-imbalance issue. GANs are widely used in the field of image processing. Exploration of GANs on insurance fraud is a unique contribution in this work. We also worked on the relative performance of other data-imbalance techniques in comparison to GANs.