HYBRID BO GA OPTIMIZATION OF XGBOOST FOR CALIBRATED AND ROBUST HYPERTENSION PREDICTION: A MULTI-COHORT VALIDATION STUDY

Background. Hypertension affects 1.3 billion people worldwide and remains the leading modifiable cause of cardiovascular mortality. Although machine learning models such as XG Boost demonstrate promising discriminative performance, their probabilistic predictions often suffer from poor calibration and limited cross population robustness, hindering large-scale clinical deployment.

Objective. To develop and validate a hybrid optimization framework combining Bayesian Optimization and a multi-objective genetic algorithm to simultaneously improve discrimination, calibration, and generalizability in hypertension prediction models.

Conclusion. The BO-GA-XG Boost framework represents a significant advancement in clinical hypertension prediction, simultaneously providing high discrimination, reliable calibration, and strong cross-population robustness three essential conditions for real-world clinical integration.