Using UMAP for Partially Synthetic Healthcare Tabular Data Generation and Validation.

Area of Science:

• Health Informatics
• Data Science
• Medical Data Generation

Background:

• Healthcare generates vast sensitive data from sensors for monitoring and diagnosis.
• Data privacy, resource intensity, and missing information due to errors necessitate new methods.
• Existing challenges include data imputation and partial data generation for incomplete datasets.

Purpose of the Study:

• To introduce a novel methodology for partially synthetic tabular data generation.
• To reduce reliance on sensor measurements and ensure secure data exchange.
• To address data privacy concerns by generating realistic synthetic samples.

Main Methods:

• Utilized Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction.
• Transformed high-dimensional reference data into a reduced-dimensional space.
• Generated and validated synthetic values for incomplete datasets using the transformed space.

Main Results:

• Successfully validated the method on prostate and breast cancer datasets.
• Demonstrated effectiveness in completing and augmenting incomplete datasets.
• Showcased superior performance compared to state-of-the-art imputation techniques.

Conclusions:

• The proposed method mitigates the need for extensive sensor readings and enhances data privacy.
• Established a formal framework for understanding and solving synthetic data generation and imputation.
• Offers a dual contribution in innovative synthetic data generation and a formal problem-solving framework.