Deep Meta- Ensemble Learning with SMOTE for Heart Disease Prediction: A Hybrid Approach

Heart disease is still a global challenge in health sector as a major reason for death. American Heart Association (AHA) reported statistics in the year 2024 about 19.91 million global deaths is due to cardiovascular disease. Hence early prediction helps in diagnosis which required reliable predictive models helps in assisting early diagnosis and preventive care helps in reducing the risk of death. This study investigates Machine Learning (ML) models and gradient boosting models performance in predicting the disease accurately. The study includes the standard dataset, which is available in data repositories like UCI and Mendeley. The study evaluates base algorithms like Logistic Regression, Decision Tree, Random Forest, Support Vector classification, Naive Bayes and gradient boosting models like XG Boost, LightGBM, CatBoost with SMOTE and without SMOTE for prediction of heart disease which show importance of data balancing in machine learning. This study gives comparison analysis of traditional ML algorithms and gradient boosting algorithms and hybrid ensembled meta learning neural network framework integrating SMOTE for handling imbalance class and meta learning with Neural Networks for efficient performance for heart disease prediction. The hybrid ensembled meta-learning framework outperformed of all other traditional methods by achieving accuracy of 94.49% and AUC-ROC of 95.44%. This study highlights that ensembled classification model when integrated with deep learning model improves their performance on balanced datasets.