SYNTHETIC CARDIOVASCULAR DATA GENERATION FOR BIOMEDICAL APPLICATIONS: COMPARATIVE EVALUATION OF FIDELITY, UTILITY, AND PRIVACY

Due to data restrictions, especially in healthcare and the biomedical domain, synthetic data (SD) is a highly effective solution for addressing data-collection challenges while maintaining participants' privacy. In this study, we selected a cardiovascular disease (CVD) dataset [1], as CVDs are the leading cause of global mortality according to the World Health Organisation (WHO), accounting for 32% of deaths [2]. We use a small subset of CVD datasets to generate synthetic data using four techniques: Conditional Tabular GAN (CTGAN), Tabular Variational Autoencoder (TVAE), Copula GAN (CG), and Gaussian Copula GAN (GC-GAN). The objective is to compare the results and assess whether synthetic data might enable public testing by researchers. Additionally, we evaluate the fidelity, utility, and privacy of the generated datasets.