The Privacy-Preserving High-Dimensional Synthetic Data Generation and Evaluation in the Healthcare Domain

1. Introduction

Synthetic data, defined by the US Census Bureau as micro data records created through statistical modeling, enhances data utility without compromising privacy and confidentiality, enhancing the utility of sensitive information (Philpott, 2017). Synthetic high-dimensional data is critical to a wide range of data science applications. Its relevance stems from the difficulties connected with gathering, disseminating, and analyzing real-world high-dimensional datasets, which frequently contain sensitive or limited information. By creating synthetic data, researchers and practitioners may address privacy issues, ethical constraints, and data limits while correctly capturing the complex structures and patterns seen in high-dimensional settings. This enables the creation, testing, and improvement of algorithms and models in a variety of disciplines, including healthcare, finance, genomics, cyber security, and others, eventually driving innovation and breakthroughs in data-driven research and applications (Giuffrè & Shung, 2023). Synthetic data addresses the challenges of obtaining and using real-world high-dimensional datasets by offering an alternative that addresses several limitations associated with genuine data sources, including cyber security, ultimately fostering innovation and advancements in data-driven research and applications (Wang et al., 2024). Real-world high-dimensional datasets frequently contain sensitive information, making them difficult to distribute or access openly. Synthetic data creation enables academics to construct privacy-preserving replacements that maintain statistical features and trends while concealing sensitive information. This allows for cooperation and experimentation without compromising privacy (James et al., 2021).

Obtaining vast and diverse real-world, high-dimensional datasets can be difficult or expensive. Synthetic data provides a solution by allowing researchers to simulate numerous scenarios and data distributions, resulting in varied datasets that accurately represent the complexity of the target domain. This is especially useful in cases where legitimate data is scarce or unavailable (Xu et al., 2021). In some sectors, ethical constraints may limit the use of real-world data, particularly if it contains personal or sensitive information. Synthetic data production helps to bypass ethical limitations by developing false datasets that mimic the features of actual data, allowing researchers to undertake experiments and analyses without ethical issues (Hao et al., 2024). Developing and testing algorithms on real-world high-dimensional datasets might be difficult owing to the aforementioned concerns. Synthetic data enables controlled trials, giving researchers the ability to evaluate algorithmic performance under a variety of scenarios without depending on potentially sensitive or restricted real-world data (Hoag, 2008).

Synthetic data is a useful technique for enhancing existing datasets. By creating extra synthetic samples, researchers may increase the size and variety of their datasets, improving the resilience and generalization capabilities of machine learning models trained on these augmented datasets (Fawaz et al., 2018). Thus, synthetic data is a diverse and powerful resource that strikes a balance between the requirement for genuine, high-dimensional data and the difficulties involved with getting and utilizing actual datasets. It promotes data science innovation, research, and experimentation while adhering to privacy and ethical guidelines.

Synthetic data, in various forms like textual, media-based, and tabular formats, is used in various domains and applications. Machine learning models enhance natural language understanding systems from synthetic data. Synthetic data overcomes privacy concerns, enables robust training, and facilitates resource-intensive applications.