Privacy through Synthesis: Exploration of Tabular Synthetic Data Generation Methods and Quality Assessment

Creating synthetic data has become a beneficial method for overcoming obstacles in data-driven applications. Synthetic data is created artificially to mimic the statistical characteristics and real-world data features without using the original sensitive personal information. Utilizing synthetic data can help to get around the problem of enough real-world data unavailability. Machine learning (ML) models can learn effectively when trained with ample amounts of data. The synthetic data provides a secure framework for testing and validating models without exposure to any personal information. It becomes a privacy-enhancing technique when it is used to create non-personal data while maintaining the same utility as real personal data. Synthetic data generation is an encouraging way for researchers in sensitive domains, such as healthcare and finance, to train their ML models, as data in these domains is not available publicly. This study aims to focus on synthetic tabular data generation techniques and evaluation metrics used for the assessment of generated tabular data.