A novel and fully automated platform for synthetic tabular data generation and validation.

Area of Science:

• Health Informatics
• Machine Learning
• Data Science

Background:

• Healthcare data access for machine learning (ML) is restricted by regulations.
• Synthetic data offers a viable alternative to real patient data.
• Existing methods for synthetic data generation require significant validation.

Purpose of the Study:

• To introduce a fully automated synthetic tabular neural generator (STNG).
• To validate and compare synthetic datasets generated by different methods.
• To enhance the accessibility of validated synthetic healthcare data for ML.

Main Methods:

• Development of STNG, integrating multiple synthetic data generators.
• Incorporation of an Automated Machine Learning (Auto-ML) module for validation and comparison.
• Empirical evaluation using twelve diverse healthcare datasets.

Main Results:

• STNG demonstrated robustness across various datasets.
• The Auto-ML module effectively validated and compared synthetic data.
• The study confirmed STNG's capability to generate high-quality synthetic healthcare data.

Conclusions:

• STNG provides a promising solution for overcoming healthcare data accessibility barriers in ML.
• Automated validation and comparison enhance trust in synthetic data.
• STNG facilitates broader ML adoption in healthcare research and practice.