<![CDATA[Heart disease remains a leading global cause of mortality, highlighting the need for accurate early risk classification. This study benchmarks Random Forest, XGBoost, and Logistic Regression for heart disease risk classification using a synthetic, perfectly balanced dataset, while addressing performance limitations caused by inadequate hyperparameter configuration. The dataset comprised 70,000 samples with a 50/50 class distribution and 18 clinical and demographic features. Although useful for controlled benchmarking, synthetic balanced data may yield optimistic estimates and may not fully represent real-world clinical variability. Each model was implemented in a scikit-learn Pipeline with median imputation and, where applicable, standard scaling. A two-stage tuning strategy was applied by combining RandomizedSearchCV with GridSearchCV refinement to optimize model configurations systematically. Under these benchmarking conditions, XGBoost achieved the best test performance, with an F1-score of 99.34%, AUC-ROC of 99.97%, and accuracy of 99.34%. Random Forest obtained an F1-score of 99.20% and AUC-ROC of 99.95%, while Logistic Regression achieved an F1-score of 99.12% and AUC-ROC of 99.95%. Age, pain in the arms/jaw/back, and cold sweats/nausea were the most influential predictors. The proposed framework is reproducible, computationally efficient, and suitable for validation on heterogeneous clinical datasets.]]>
Copyrights © 2026
