AI-Based Explainable Hybrid Model for Early Prediction of Diabetes

Early prediction of diabetes can significantly reduce long-term health complications and healthcare costs. This study proposes an explainable artificial intelligence (AI) framework that integrates Extreme Gradient Boosting (XGBoost) with SHapley Additive exPlanations (SHAP) to predict diabetes risk using clinical and lifestyle data. Publicly available datasets, including the PIMA Indian Diabetes and Kaggle Diabetes datasets, were preprocessed and analyzed to develop a transparent predictive model capable of balancing accuracy and interpretability. The proposed hybrid model achieved an AUC of 0.81, outperforming traditional machine-learning baselines. The SHAPbased interpretability layer identified key predictors such as glucose level, BMI, age, and blood pressure, enabling clinicians to understand how individual risk factors contribute to predictions. This research demonstrates that explainable AI models can support trustworthy, data-driven preventive healthcare and inform early clinical decision-making in diabetes management.