CardioEnsemNet, an optimized ensemble learning framework, predicted coronary artery disease across four benchmark datasets with accuracy exceeding 97%, F1-scores >95%, and ROC-AUC close to 0.99.
Does the CardioEnsemNet ensemble learning framework accurately predict coronary artery disease across diverse clinical datasets?
The CardioEnsemNet machine learning framework demonstrates high accuracy and interpretability for predicting coronary artery disease across multiple heterogeneous datasets.
One of the major causes of mortality around the world, poor diagnosis and the early stages of the disease being asymptomatic are the primary reasons for the occurrence of coronary artery disease (CAD). Therefore, early detection of CAD is essential to decrease the risk of death and improve clinical intervention. Machine learning (ML) methods have shown good performance for CAD prediction; however, existing methods face challenges such as limited cross-dataset generalizability, suboptimal feature selection, poor performance under class imbalance, and limited interpretability. To overcome these drawbacks, this study proposes a leakage-safe ensemble learning framework, namely CardioEnsemNet, for accurate and interpretable CAD prediction from heterogeneous clinical datasets. The framework is based on a combination of dataset-aware preprocessing, advanced feature engineering techniques, including polynomial and interaction features, Boruta feature selection, and SMOTE class balancing. To enhance predictive performance, the optimised stacking ensemble architecture is used, combining classifiers such as Random Forest, Support Vector Machine, K-Nearest Neighbours, XGBoost, and a meta-learner logistic regression. Furthermore, SHAP and LIME are added to provide interpretability of model predictions at both global and local scales. Four benchmark datasets, UCI Cleveland Heart Disease, Statlog Heart, Z-Alizadeh Sani and Framingham Heart Study were used to evaluate the proposed framework. Experimental results show that predictive performance is consistently high, with accuracy values exceeding 97%, F1-scores exceeding 95%, and ROC-AUC values close to 0.99 across all datasets. The robustness and contribution of the proposed preprocessing and ensemble learning strategies are then validated and confirmed by using cross-dataset validation and ablation studies. The results indicate that the CardioEnsemNet is a reliable, interpretable model for predicting CAD across diverse clinical data conditions.
Aswani et al. (Sun,) conducted a other in Coronary Artery Disease. CardioEnsemNet was evaluated on Coronary artery disease prediction accuracy. CardioEnsemNet, an optimized ensemble learning framework, predicted coronary artery disease across four benchmark datasets with accuracy exceeding 97%, F1-scores >95%, and ROC-AUC close to 0.99.