Implementation of XGBoost for diabetes mellitus risk prediction based on health history

Abstract

Diabetes mellitus (DM) is a chronic disease with a growing global burden and specific challenges for early management, particularly in regions with limited access to healthcare. This study develops a web-based system to classify diabetes risk from medical history using extreme gradient boosting (XGBoost), an ensemble model of decision trees. The dataset comprised 520 respondents (320 DM, 200 non-DM) and underwent labeling, standardization, and an 80:20 train–test split, followed by hyperparameter selection via grid search and 5-fold cross-validation (CV). On the test set, the model achieved an accuracy of 0.9888, precision of 1.0000, recall of 0.9718, and an F1-score of 0.9857; discriminative performance was also strong with an area under the receiver operating characteristic curve (AUC ROC) of 0.839. These findings confirm that XGBoost effectively handles complex or imbalanced medical data while providing probabilistic outputs that are clinically meaningful. Deployed as a web application, the system can support early screening, triage, and clinical decision-making, thereby expediting referrals and personalizing interventions in primary care and hospital settings, especially in resource-constrained environments. This work lays the groundwork for further development, including the integration of explainable artificial intelligence (XAI) techniques to enhance clinical transparency.