Heartbeat classification using projected and dynamic features of ECG signal

Abstract A novel method for the electrocardiogram (ECG) beat classification according to a combination of projected and dynamic features is presented. Projected features are derived from a random projection matrix, in which each column is normalized and each row is transformed by discrete cosine transform (DCT). In addition, three weighted RR intervals are determined as the dynamic features. A support vector machine classifier is used to cluster heartbeats into one of 15 or 5 classes by using those two kinds of features. The raised method acquires an overall accuracy of 98.46% in the “class-based” assessment strategy and 93.1% in the “subject-based” assessment strategy, based on the MIT-BIH arrhythmia database. These results show that the raised method has better performance, compared with the state-of-the-art automated heartbeat classification systems.