Coronary heart disease prediction on small datasets: A comparative analysis

Abstract

Coronary Heart Disease (CHD) is one of the major causes of death worldwide. Bangladesh and other developing nations face similar challenges. Most people wait until it is too late to recognise that their cardiac problems are getting worse. For this reason, early detection is essential to reduce the death toll or major health effects from CHD. This paper's main goal is to use supervised machine learning (ML) techniques to improve the accuracy of CHD prediction for a Bangladeshi population. ML methods including KNN, Random Forest, Decision Tree, Naive Bayes, and Binary Logistic Regression Model are used in our research methodology to predict CHD on two distinct datasets: one from Bangladesh and the other from Canada. Synthetic data for Bangladeshi dataset were generated by using ADASYN which produces accuracy of 88.12%. On the other hand, using SMOTE, the obtained accuracy was around 93.79%. Both accuracies were achieved by applying Random Forest Algorithm. Binary Logistic Regression obtained highest accuracy for the Canadian dataset which is 72.33%.