Novel Deep Convolutional Neural Network based Classification of Arrhythmia

Abstract


 Holter monitors are used to record Electrocardiogram (ECG) data which is extremely hard to analyze manually. Convolutional Neural Network (CNN) are known to be efficient for classification of image data. Hence, in this study, we are using Deep Convolutional Neural Network to classify the ECG data into various types of Arrhythmias. Denoising, segmentation and data augmentation techniques are used for pre-processing of the data. The proposed model uses the MIT-BIH Arrhythmia Dataset for training and evaluation purpose this dataset has much imbalance which has been removed using data augmentation techniques. The proposed approach shows an overall accuracy 99.67% along with 99.68% precision and 99.66% recall. Further, we have also compared the state-of-the-art models like 2D CNN, genetic ensemble of classifiers, Long Short-Term Memory (LSTM) Networks, etc results with proposed model. And the introduced approach is outperforming when compared to these models.