Explainable AI for Diabetes Risk Stratification Using Severity Index Indicators

This study presents a machine learning framework designed to stratify risk and predict the severity of chronic diabetic complications. Leveraging BMI, family history of diabetes, and lifestyle factors, a novel Severity Index was developed by integrating demographic, lifestyle, and medical attributes, facilitating patient categorization into distinct risk levels. Employing advanced clustering techniques like K-Means and Hierarchical Clustering, along with a Random Forest classifier, the model achieved a better 99.03% accuracy in predicting severity levels. Model explainability via SHAP was utilized to identify significant predictors, including BMI, smoking habits, and the Severity Index, ensuring model transparency and clinical relevance. This framework highlights the potential for AI to transform chronic disease management by offering insights into patient-specific risk factors, thereby supporting early intervention and personalized care.