A comparative study of supervised learning techniques for ECG T-wave anomalies detection in a WBS context

Today, most of Wireless Body Sensors (WBS) for remote monitoring of cardiovascular disease, rarely include automatic analysis and detection of ECG abnormalities, or are limited to cardiac arrhythmia´s. The detection of more complex cardiac anomalies such as Ischemia or myocardial infarction, requires an advanced analysis of ECG wave Known as P, Q, R, S, and T, especially the T-wave, which is often associated with serious cardiac anomalies. The goal of this paper is to study the classification of T-wave abnormalities with consideration to a context of wireless monitoring system. The study approach is based on experimentation and comparison of classification performance and response time of 7 supervised learning models. We performed our experiments on a real ECG data from the EDB medical database from Physionet. Our results show that the decision trees models offer better results with, on average, an Accuracy of 92.54 %, a Sensitivity of 96.06%, a Specificity of 55.41% and an Error Rate 7.41%.