Heart Murmurs Clustering Using Machine Learning

This paper presents an approach of utilizing machine learning to assist cardiac auscultation for heart murmur detection. Our aims are to compare the effectiveness and advantages of both supervised and unsupervised techniques in performing clustering and classification of selected sets of different heart murmur episodes. The supervised approach uses a decision tree while the unsupervised method uses as few parameters as possible and can also reveal the most effective parameters in clustering. The clustering results are then compared and the boons and banes are drawn. The supervised approach implementation is straightforward following the known diagnosis procedure with little training. The caveat, however, is that it fails to detect extreme anomalies. The unsupervised clustering approach needs more computation and training with the advantage of self-learning and revealing most effective features in clustering the murmur episodes caused by different cardiovascular anomalies.