Predicting Coronary Heart Disease Mortality Using Machine Learning Algorithms

This study aimed to predict Coronary Heart Disease (CHD) mortality over a 10 to 15-year period using Machine Learning (ML) techniques. We studied a rather large variety of features associated to CHD mortality from the Sleep Heart Health Study (SHHS) dataset. Using the most important features identified through mutual information, several ML algorithms, including LR, SVM, KNN, ETC, and RF, were trained. A 10-fold cross-validation was employed to ensure accuracy. As a result, the KNN algorithm (k = 18) outperformed the others, achieving an accuracy of 76.05%, an AUC of 79.41%, a sensitivity of 81.08%, a specificity of 71.28%, and a precision of 73.48%. Our proposed approach reasonably well predicts CHD mortality many years in advance, aiming to identify critical factors for CHD mortality prediction, monitor patients, and provide early warnings of the need for prevention and treatment through medical intervention.