Equitable AI for Early Detection: A Fairness-Aware Machine Learning Model for Pediatric Type 2 Diabetes Risk Prediction in Underserved US Populations Using Multi-cycle NHANES Data

The premature development of type 2 diabetes in children in the United States is disproportionately concentrated among socioeconomically disadvantaged and underserved populations, underscoring the need for equitable, evidence-based prevention strategies. This study presents a machine learning model to predict early risk of pediatric type 2 diabetes using multi-cycle data from the National Health and Nutrition Examination Survey (NHANES) spanning 2013–2018. The model integrates clinical indicators, behavioral factors, and social determinants of health to evaluate predictive performance across racial, ethnic, and socioeconomic subgroups. Fairness was assessed using equity-sensitive metrics, including demographic parity and equalized odds, alongside traditional performance measures. Results demonstrate that the model achieves strong predictive accuracy while maintaining consistent performance across subgroups, indicating reduced disparity in risk prediction without compromising clinical utility. These findings highlight the potential role of equity-focused evaluation frameworks in supporting early identification of pediatric diabetes risk and informing public health screening efforts in underserved U.S. communities.