Optimizing hypertension prediction using ensemble learning approaches

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Cardiovascular Disease Prediction

Background:

• Accurate hypertension (HTN) prediction is vital for preventive healthcare.
• Traditional single-model methods often lack optimal predictive accuracy for HTN.
• Ensemble learning offers a promising approach to enhance predictive performance.

Purpose of the Study:

• To evaluate the efficacy of ensemble learning techniques for improving hypertension prediction accuracy.
• To identify key risk factors associated with hypertension using advanced feature selection and explainability methods.
• To compare the performance of various machine learning models, including a stacking ensemble, for HTN prediction.

Main Methods:

• Utilized a dataset of 612 Ethiopian participants with 27 potential HTN risk features.
• Employed multi-faceted feature selection (Boruta, Lasso, Fwd/Bwd, RF importance) to identify 13 common features.
• Trained and evaluated logistic regression, ANN, RF, XGB, LGBM, and a stacking ensemble model using accuracy, precision, recall, F1-score, and AUC.
• Applied SHapley Additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• The stacking ensemble model achieved the highest performance: 96.32% accuracy, 95.48% precision, 97.51% recall, 96.48% F1-score, and 0.971 AUC.
• SHAP analysis identified weight, drinking habits, hypertension history, salt intake, age, diabetes, BMI, and fat intake as significant HTN risk factors.
• Ensemble learning demonstrated superior predictive accuracy and robustness compared to single models.

Conclusions:

• Ensemble learning, particularly stacking, significantly enhances hypertension prediction accuracy and robustness.
• Key modifiable risk factors like weight, lifestyle choices, and dietary habits are critical for HTN prediction.
• This research supports the integration of advanced ML techniques for optimized early intervention and personalized hypertension management.