<![CDATA[Hypertension remains a major contributor to cardiovascular morbidity, yet community screening often relies on late or opportunistic measurements. This study compares Logistic Regression, Random Forest, Gradient Boosting, XGBoost, and Support Vector Machine for hypertension risk prediction using a public dataset of 1,985 patient records and ten demographic, clinical, and lifestyle predictors. Numerical features were standardized, categorical features were one-hot encoded, and missing medication values were imputed using the mode. Models were evaluated using an 80:20 stratified train-test split and 5-fold cross-validation. The best model was optimized with RandomizedSearchCV and examined through feature importance and ablation study. Tuned Gradient Boosting achieved the highest cross-validation F1-score of 0.9994 and ROC-AUC of 1.0000. On 397 test records, it obtained 0.9975 accuracy, 1.0000 precision, 0.9951 recall, 0.9976 F1-score, and 1.0000 ROC-AUC, with only one false negative. Ablation showed that removing blood pressure history reduced test F1-score to 0.7681, confirming its dominant contribution. The model is promising for screening support, but external clinical validation is required before diagnostic use.]]>
Copyrights © 2026
