Motion artifacts recognition in electrocardiographic signals through artificial neural networks and support vector machines for personalized health monitoring

Abstract

Nowadays a personalized approach is being giving to health care concerning the prevention of diseases, improving diagnosis and treatment of patients, for this, equipment to measure ambulatory vital signs are used, allowing to get large volumes of information. Nevertheless, the obtained information from ambulatory electrocardiography has no largely clinical validity because it is contaminated with motion artifacts, for this reason, it is necessary to determine what information is useful and what information can be ruled out. This paper presents a comparison between two different classification methods of electrocardiography signals: Artificial Neural Networks and Support Vector Machines. Database includes electrocardiography signals of volunteers and some important features of these signals are extracted to train both classification methods. Also, performance of methods is assessed verifying the generalization capabilities. The best performance was presented by the Radial Basis Function kernel.