Cardiovascular disease continues to be a primary cause of death globally, requiring precise and clinically interpretable predictive models. This paper presents an innovative IoT-enabled hybrid framework for predicting heart disease, incorporating advanced data balancing, feature optimization, ensemble learning, and explainable artificial intelligence (XAI). The proposed model utilizes SMOTE-ENN to rectify class imbalance, succeeded by a two-tier feature selection process that integrates Mutual Information and SHAP-based importance ranking. A heterogeneous ensemble comprising XGBoost, LightGBM, Random Forest, CatBoost, and TabNet classifiers is constructed, with their probabilistic outputs integrated through a logistic regression meta-learner. Comprehensive experiments performed on the UCI Heart Disease dataset and real time database collected from hospital demonstrate that the proposed method attains exceptional performance, with accuracy at 98.56%, sensitivity at 97.5%, specificity at 97%, AUC at 0.99, and Matthews Correlation Coefficient at 0.94 with 5-fold cross validation. Moreover, SHAP analysis offers clear global and local explanations, thereby improving clinical trust and interpretability. The findings validate that the suggested framework is both resilient and appropriate for practical healthcare decision-support systems.
Keywords: Heart Disease, SMOTE-ENN, CatBoost & TabNet, Machine Learning, Tree-Importance, Explainable AI, SHAP, and Stacked Ensemble learning.
How to cite this article: Singh R, Naz S, Payal H, Singh H, An IoT-Enabled Explainable Stacked Ensemble Framework with SMOTE-ENN for Robust Heart Disease Prediction. Int J Drug Deliv Technol. 2026;16(4s): 974-987; DOI: 10.25258/ijddt.16.4s.115
Source of support: Nil
Conflict of interest: None