Using GANs (Generative Adversarial Networks) to generate fake patients

Abstract

This master thesis is a continuation of the investigation line opened in with the Generative Adversarial Network based Machine for Fake Data Generation thesis. The objective is using Generative Adversarial Networks (GANs) for the generation of fake data for anonymizing patients information in the health sector. This way, new samples can be generated without putting personal data at risk and, if the quality is good enough, can be used in educational and research areas. A systematic revision of the previous works has been made, and a line of work has been developed in the area of electrocardiograms, using the techniques that have given better results in other papers. Several GAN models have been made, each of them trying different features, and satisfactory results have been obtained in some of them. Moreover, quantitative measuring techniques have been included and tested.