Harnessing Artificial Intelligence for Heart Disease: Comparing Machine Learning, Deep Learning, and Bayesian Network Approaches

Heart disease is still a global health challenge; thus, the model at an early stage of prediction in a timely intervention remains a challenge. It compares various techniques in machine learning, deep learning, and Bayesian networks against the UCI Cleveland Heart Disease dataset, whereby it compares with the Kaggle Heart Disease dataset. Attempts have been made for the following models: XGBoost, Random Forest (RF), Gradient Boosting, Logistic Regression (LR), Support Vector Classifier (SVM), Convolutional Neural Networks (CNN), and Bayesian Networks. One-hot encoding, ANOVA F-test-based feature selection, and SMOTE will be used for advanced preprocessing for higher performance. In ensemble methods, XGBoost was the best method with accuracy of about 97.32% while the second highest accuracy rate was given by Random Forest with about 95.89%. This result probably justifies the applicability of ensemble models in medical decisions. A very high performance was observed while using a measure that is derived from precision, recall, and F1-score. The paper puts into the limelight the necessity for feature engineering, missing data correction, and class imbalance correction for higher predictive reliability. Future research should be placed on DL with a more deep architecture and multiple data sets for the practical application in healthcare with improved prediction accuracy.