Generative Adversarial Networks for Synthetic Data Generation in Privacy-Preserving Health Informatics

Privacy preserving health informatics deals with the problem of how to obtain informative analysis from the health-related information while ensuring that the privacy of individuals is protected. This paper presents the first framework that employs GANs for synthetic data generation in the context of health informatics to overcome the presented shortcomings and expand on existing literature. In this framework, the solution is presented through the utilization of GANs which presents a nucleation point in developing synthetic datasets that reproduce the density and distribution of real patient data. Employing PLMs in conjunction with synthetic as well as original datasets, led to accurate classification rates of above 98%. These results raise the possibility of the synthetic data generated by model substituting the original data for machine learning purposed while maintaining high accuracy and, arguably most importantly, privacy of the individual in question. Therefore, our work presents a major advance toward bridging the paradigm of health data usability and protection.