Evaluation of Machine Learning Techniques for ECG T-Wave Alternans

Abstract

This paper presents the evaluation of T-wave alternans (TWA) detection based on machine learning techniques. F1-score metric was used for optimal features set selection. KNN, LR, RFC, SVC classifiers were evaluated as a part of TWA detection system. T-criteria was proposed as justification approach to select optimal features number based on classifiers performance. Designed optimal TWA classification system was evaluated on PhysioBank database and classification F1-score about 89% was obtained for RFC classifier.