Towards the Design, Quality Assessment and Explainability of Synthetic Tabular Data Generation Techniques for Metabolic Syndrome Diagnosis

In last years decision-making Machine Leaning (ML) approaches have evolved from traditional methods to evidence-based approaches, particularly in healthcare sector. However, sharing data with third parties raises significant security and privacy concerns. To address these issues, researchers have explored data anonymization, distributed privacypreserving data mining, and synthetic data generation (SDG). SDG, in particular, shows promise in enabling secure data sharing while preserving privacy, crucial for developing advanced AI models. This paper focuses on Metabolic Syndrome (MetS) data, a condition affecting a significant portion of the population, and investigates various synthetic tabular data generation (STDG)techniques. It evaluates the performance of an AutoML approach for predicting MetS using different percentages of synthetic data assessed through a specific evaluation framework. Moreover, presents an explainability and feature relevance analysis of the proposed STDG methods.