Advanced Heart Disease Detection Using Stacked Machine Learning Classifiers

Abstract

Background: Cardiovascular disease remains a primary global health concern, highlighting the urgent need for reliable and widely applicable diagnostic tools. This work aims to present an Advanced Heart Disease Detection system that integrates clinical data from four major international datasets: the Cleveland, Hungarian, Switzerland, Long Beach VA datasets, and Statlog Heart Diseases for CAD and No-CAD diseases. The system utilizes 14 clinically-relevant standardized features and employs an automated preprocessing pipeline that includes median imputation for numeric data, “most frequent” imputation for categorical attributes, and standard scaling for data integrity.  

Methods: The proposed methodology is implemented in two phases. The first phase is learned with Deep Q-Network, which achieved high accuracy compared with others. The second phase is a Stacked Ensemble voting algorithm, which is synthesized by a Logistic Regression, Support Vector Machine (SVM), XGBoost (XGB), Multilayer Perceptron, and Random Forest (RF). The designed model has increased the performance accuracy by using extensive hyperparameter tuning with GridSearchCV. The results show high accuracy compared with other studies.

Results: The first model achieved 92.65%, while the second model achieved 92.52% % for DQN and ensemble methods, respectively. The proposed model demonstrated high and reliable performance, indicating its strong and reliable tool for early detection of heart diseases, partially for classification into CAD and No CAD and its applicability in real-world clinical decision support systems.