A Cost-Aware Stacking Ensemble Framework for Early Heart Disease Detection

Heart disease remains a leading cause of mortality
worldwide, emphasizing the need for reliable and early
screening mechanisms. Although machine learning techniques
have been widely explored for heart disease prediction, many
existing approaches primarily optimize overall accuracy and
provide limited attention to misclassification cost asymmetry
and interpretability, which are critical for clinical screening
applications. This paper proposes a cost-aware stacking
ensemble framework for early heart disease detection that
integrates heterogeneous base classifiers, optimized decision
thresholding, and explainable artificial intelligence techniques
within a unified pipeline. Logistic regression, support vector
machine, and random forest models are employed as base
learners, and their outputs are combined using a logistic
regression–based meta-learner. Cost-sensitive learning is
incorporated to prioritize reducing false negatives, while the
final classification threshold is optimized using receiver
operating characteristic analysis. Model transparency is
enhanced through Shapley value–based local explanations and
permutation-based global feature importance analysis.
Experimental evaluation on the UCI Heart Disease dataset using
stratified cross-validation demonstrates that the proposed
framework achieves an accuracy of 87.17%, a sensitivity of 0.92,
a specificity of 0.84, and an area under the ROC curve of 0.87.
The results indicate that the proposed approach provides a
robust and interpretable decision-support system suitable for
early heart disease screening.