Enhancing Sleep Disorder Prediction Through Feature Engineering and Stacking Ensemble Learning on Imbalanced Lifestyle Data

Abstract

Undiagnosed sleep disorders pose significant cardiovascular risks, necessitating accessible screening tools beyond invasive clinical procedures. This study aims to develop a robust diagnostic framework using the Sleep Health and Lifestyle Dataset. To address class imbalance and enhance predictive sensitivity, a Stacking Ensemble architecture integrating Random Forest, Gradient Boosting, CatBoost, and XGBoost is implemented, augmented by Pulse Pressure feature engineering and the Synthetic Minority Over-sampling Technique (SMOTE). The proposed model achieved a superior accuracy of 98.61% and a recall of 99.24%, significantly outperforming single classifiers. Feature analysis further identified heart rate and sleep duration as critical physiological determinants. These findings conclude that combining feature engineering with optimized ensemble learning offers a highly accurate diagnostic approach with rapid training convergence, providing a scalable pathway for early sleep disorder detection.