Electrocardiogram signal classification to detect arrythmia with improved features

Arrhythmia detection and classification plays a significant role in a timely management of cardiac issues and thus help in a sustained life. In this paper, a new scheme for classification of all type of arrhythmias including those that are life threatening as well as those which are not dangerous for life is presented. This scheme is based on a new set of features including Discrete Wavelet Transform (DWT) as morphological feature, whereas, RR-interval and Teager energy operator (TEO) are used as dynamic features to represent the heartbeats. Independent Component Analysis (ICA) is also applied on DWT coefficients as a dimensionality reduction technique to balance the importance of two types of features used. All these features are combined and concatenated to finally classify heartbeats into eighteen classes by using a neural network based classifier. This method is tested and validated over entire dataset of MIT-BIH arrhythmia dataset, resulting in an overall accuracy of 99.4%.