A stacked ensemble machine learning framework for healthcare risk prediction

Accurate prediction of patient health risk is important for early diagnosis and clinical decision-making. However, healthcare data is often complex and noisy, which limits the performance of individual machine learning models. This study proposes a stacked ensemble framework for healthcare risk prediction. The approach combines Random Forest, LightGBM, XGBoost, and CatBoost as base learners, with Logistic Regression as a meta-learner. The model is trained on 49,942 patient records with 15 clinical and lifestyle features. Class imbalance is handled using SMOTE applied within a stratified 5-fold cross-validation framework. SHAP-based explainability is used to improve interpretability of model predictions along with preprocessing steps such as feature scaling and encoding. Experimental results show strong performance across all evaluation metrics, outperforming individual models.