X-DiaRisk: Explainable AI based Diabetes Risk Prediction from Clinical and Lifestyle Data
TL;DR
The interpretability analysis revealed that family history of diabetes, physical activity, and blood pressure are the most influential predictors, and the proposed X-DiaRisk framework offers a practical tool for physicians, supporting efficient screening and early identification of individuals at risk of diabetes, thereby combining high predictive accuracy with explainability.
Abstract
Diabetes risk has become one of the most acute concerns in modern healthcare. The rising global prevalence of the disease increases the need for early intervention. The proposed X-DiaRisk framework is a comprehensive solution of automated risk prediction of diabetes based on systematized healthcare information. The proposed framework employs an Optuna-based CatBoost model as its baseline model along with LIME. Experimental analysis shows higher predictive performance with R2 of 0.8498, RMSE of 3.5206 and MAE of 2.8307. To increase the transparency, both LIME and SHAP are employed, providing local and global insights. The interpretability analysis revealed that family history of diabetes, physical activity, and blood pressure are the most influential predictors. A comparative evaluation between LIME and SHAP demonstrated strong agreement, with a Spearman correlation of 0.70 confirming the robustness of explanations. Thus, the proposed X-DiaRisk framework offers a practical tool for physicians, supporting efficient screening and early identification of individuals at risk of diabetes, thereby combining high predictive accuracy with explainability.