Paper Title
Classification of Ecg Beats for Types of Arrhythmia Using Cascaded Lstm and Cnn
Abstract
The work presented in this paper aims to classify ECG beats into five super classes of Arrhythmia. The learning model is trained using a cascaded model of long short term memory network and Convolution neural network. The data for classification is acquired from the Arrhythmia database provided by the Massachusetts Institute’s Beth-Israel hospital. The data is labeled using the help of the annotations of the Cardiologists. The features chosen are the local maximas of a segmented beats. The signal records of 10 sec duration are then segmented into beats of 100 samples each. The beats are compiled as a training dataset. A training and validation split of 80-20% is made upon which the performance metrics of the classifier are evaluated. The maximum classification accuracy obtained is 98.2% From the confusion matrix it can be deduced that average precision and average recall per class is obtained to be 96.94% and 97.12% respectively.
Keywords - Machine Learning, Lstm, Convolution, Back Propagation, Activation Function, Svm, Accuracy