Review on Heart Disease Prediction using Machine Learning Approaches

Abstract

In this paper, the five machine learning models of Random Forest, Support Vector Machine, Logistic Regression, XGBoost, and K-Nearest Neighbors are carefully compared and evaluated for the prediction of heart disease based on a clinical dataset. The objective is to describe the accuracy, strengths, weaknesses, and applicability of each model with emphasis on which one among the algorithms works best for heart disease diagnosis. We mainly evaluated them in terms of accuracy: Random Forest was the highest among all models, with an accuracy of 89%. It showed good precision and recall abilities owing to being based on ensemble learning. Closely followed was XGBoost with an accuracy of 85% but much more computationally intense, which meant it needed more machine learning processes to compensate for complex clicked patterns. Logistic regression obtained an accuracy of 81%, providing good confidence in recalling positive cases and therefore can identify true-positive instances well. Moderate accuracy at 74% is reported for SVM but its computational intensity and sensitivity to parameter tuning impeded performance. The KNN, with its straightforward design, had lower overall performance, at only 69% accuracy, given that it struggles on feature scaling and is sensitive to irrelevant information in the dataset. Overall, Random Forest and XGBoost have shown the greatest promise in predicting heart disease. Further tuning on these models would enhance their predictive performance in the pursuit of prediction applications.