-
1
The study pooled three publicly available heart disease datasets from Kaggle, UCI, and Cleveland to predict cardiovascular disease.
-
2
The authors used Instance Hardness Thresholding and Tomek Links to address class imbalance and noisy class boundaries.
-
3
Scaling, balancing, and nested hyperparameter tuning were performed within training folds, with no preprocessing applied to held-out test data.
-
4
The study compared 12 classifiers and combined top-performing models using soft- and hard-voting ensembles.
-
5
Evaluation included calibration, decision curve analysis, threshold analysis, SHAP explanations, and feature ablation; the authors noted record overlap and no external validation.
