Evaluation of Synthetic Data Generation Models for Balancing Multiclass Metabolomic Profiles

Balancing highly imbalanced multiclass datasets, a crucial step for effective classification in health and well-being analysis, remains challenging, especially with conditions like Metabolic Syndrome (MetS). This study evaluates Synthetic Data Generation (SDG) techniques for balancing a dataset of metabolomic profiles associated with MetS. Evaluating models like Gaussian Copula, FAST ML, CTGAN, TVAE, TabDDPM, and Tabsyn reveals their effectiveness in preserving multivariate relationships but dissimilar success in replicating individual variable distributions. Results show significant improvements in classification tasks with balanced data, highlighting the potential of SDG in enhancing real-world data analysis and decision-making in health research.