The Study of Algorithms of Machine Learning Used for Classification and Detection of Cardiac Disease, and Their Early Detection

Abstract

This paper presents machine learning algorithms to predict cardiac disease based on a curated literature corpus of clinical and signal-processing studies. The dataset summarizes performance metrics of different traditional classifiers (Logistic Regression, k-Nearest Neighbors, Naive Bayes, Decision Tree, Support Vector Machine, Random Forest, Gradient Boosting/XGBoost, Extra Trees) and deep learning architectures (Artificial Neural Networks (ANN), Convolutional Neural networks(CNN), Recurrent Neural networks(RNN), hybrid CNN–LSTM models) on various heart disease, cardiovascular disease (CVD) and ischemia detection tasks. We mined the literature for algorithm–metric pairs (mostly accuracy) and related cardiac disease types, and normalised the data to facilitate cross-study comparison. We extracted and normalised algorithm-metric pairs (mostly accuracy) and related cardiac disease types for cross-study comparisons. Tree-based ensemble methods (Random Forest, Gradient Boosting, XGBoost, Extra Trees) and CNN-based models typically report higher accuracies on small, well-curated tabular and phonocardiogram/ECG datasets with 60+ reports exceeding 95–99%, whereas population cohort performances (e.g., BRFSS) have been more modest (~70–80%) over a larger unbalanced dataset. Random Forest and XGBoost turn out to be strong top performers for tabular clinical data, whereas CNN architectures dominate signal-based ischemia and cardiac sound classification. Overall, the results indicate that CNNs and ensemble tree models are currently the most promising algorithmic families for the diagnosis of cardiac illness; however, before widespread clinical deployment, more work on evaluation metrics, generalisability, and multi-center validation is needed.