Synthetic Data Generation for Pediatric Diabetes Research Using GANs and WGANs
Universidad Autónoma de Zacatecas, Antonio García-Domínguez, Carlos Eric Galvan-Tejada, Universidad Autónoma de Zacatecas, Rafael Magallanes‐Quintanar, Universidad Autónoma de Zacatecas, Miguel Cruz, Centro Médico Nacional Siglo XXI, Miguel Alexander Vázquez‐Moreno, Centro Médico Nacional Siglo XXI, Erika Acosta-Cruz, Universidad Autónoma de Coahuila · Revista Mexicana de Ingeniería Biomédica · 2025
Pediatric diabetes research is often constrained by data scarcity, hindering the development of accurate predictive models for clinical applications. This study addresses this limitation by evaluating the effectiveness of Generative Adversarial Networks (GANs) and Wasserstein GANs (WGANs) in generating synthetic datasets that replicate the statistical properties of real pediatric diabetes data. A structured methodology was applied, incorporating preprocessing, model design, and dual evaluation metrics: Jensen-Shannon and Kullback-Leibler divergences for statistical fidelity, and a classification model to assess practical utility. Results demonstrate that both models produce high-fidelity synthetic datasets, with WGANs showing superior performance in capturing complex patterns due to improved training stability. Nonetheless, challenges remain in replicating the inherent variability of pediatric data, influenced by growth and developmental factors. This work highlights the potential of synthetic data to augment pediatric diabetes datasets, facilitating the development of robust and generalizable predictive models. Limitations include the dependency on initial data quality and the specificity of the models to pediatric datasets. By addressing critical gaps in data availability, this study contributes to advancing AI-driven healthcare solutions in pediatric diabetes research.