Explainable Ensemble XGBoost Model for Diabetic Peripheral Neuropathy Risk Prediction with Android Deployment

Abstract

Diabetic Peripheral Neuropathy (DPN) is a common chronic complication of diabetes mellitus that can reduce quality of life and increase the risk of ulcers and amputation if not detected early. Limited access to clinical screening tools highlights the need for accessible preventive risk assessment systems, while existing machine learning approaches have primarily focused on improving predictive performance with limited attention to model interpretability and practical deployment in real-world clinical settings. This study proposes an explainable weighted ensemble XGBoost model integrated into an Android-based mobile application for early DPN risk prediction using the Diabetes 130-US Hospitals dataset. The proposed framework employs Hyperopt to optimize three complementary XGBoost models, whose predictions are combined through validation-based weighted averaging according to their R² performance. The proposed model achieved strong predictive performance, with an RMSE of 0.0189, MAE of 0.0122, MAPE of 7.5234%, and R² of 0.7269. SHAP analysis identified key risk factors, including insulin usage, Body Mass Index (BMI), and HbA1c levels, highlighting the importance of metabolic control in DPN progression. The trained model was deployed in an Android application using ONNX Runtime to enable offline prediction. User Acceptance Testing demonstrated a satisfaction rate of 90.56%, indicating good usability. The results demonstrate that the proposed system provides an interpretable and practical decision-support tool for early DPN risk assessment while improving the accessibility of mobile-based screening solutions.