Evaluation of machine learning methods for the long-term prediction of cardiac diseases

Abstract

We evaluate several machine learning algorithms in the context of long-term prediction of cardiac diseases. Results from applying K Nearest Neighbors Classifiers (KNN), Support Vector Machines (SVM) and Random Forests (RF) to data from a cardiological long-term study suggests that multivariate methods can significantly improve classification results. SVMs were found to yield the best results in Matthews Correlation Coefficient and are most stable with respect to a varying number of features.