Explainable AI Framework for Precision Public Health in Metabolic Disorders: A Federated, Multi-Modal Predictive Modelling Approach for Early Detection and Intervention of Type 2 Diabetes

Abstract

One of the biggest public health problems of the twenty-first century is metabolic disorders, especially Type 2 diabetes (T2D). Morbidity, mortality, and medical expenses can be significantly decreased by early detection of at-risk people. However, nonlinear, multi-factorial, and high-dimensional interactions that influence the development of disease are not well captured by traditional risk-scoring methods. In order to predict and interpret the risk of type 2 diabetes and related metabolic disorders, this study creates an Explainable AI (XAI) framework for precision public health that combines multi-modal data, such as genomic profiles, lifestyle factors, socioeconomic determinants, and electronic health records (EHR). We create a federated, hybrid model that combines Random Forest classifiers, Deep Neural Networks (DNN), and Gradient Boosting Machines (LightGBM/XGBoost), building on federated and ensemble learning paradigms. Shapley Additive Explanations (SHAP) and counterfactual analysis are used to uncover personalized, actionable risk profiles in order to attain explainability. Harmonized multi-institutional datasets with over 200,000 records gathered from several U.S. health systems are used to train the model. The results show a calibrated Brier score of 0.12, sensitivity of 89%, specificity of 87%, and AUC of 0.93 ± 0.01. The socioeconomic deprivation index, polygenic risk score, BMI slope, and HbA1c trajectory are the main factors, according to SHAP study. Federated deployment protects data privacy while preserving performance. These results show that federated, explainable AI pipelines can facilitate population-based, privacy-preserving, andThe goal of precision public health is being advanced by large-scale early-warning systems for managing metabolic diseases.