Abstract

Classification of electrocardiograph (ECG) signals are an important part in clinical diagnosis of cardiovascular disease. According to World Health Organization (WHO), 17.9 million people die each year from cardiovascular diseases, accounting for an estimated 31% of all deaths worldwide.

This study will focus on ECG signal classification for Arrhythmia – a type of cardiovascular disease that refers to any change from the normal sequence of electrical impulses. Using Physionet’s MIT-BIH Arrhythmia Dataset by Fazeli, several deep learning models are developed to categorize ECG signals into five classes, as seen in previous research by Mohammed et.al. [4]: Normal (N), Supraventricular premature (S), Premature ventricular contraction (V), Fusion of ventricular and normal (F) and Unclassified (Q). Random Forests (RF) and Gradient Boosting Method (GBM) were used as baseline models with 92.5% and 92.4% test accuracies, respectively.

Moreover, three deep learning architectures were implemented, namely 1D and 2D Convolutional neural networks and a Fully-connected neural network (FC-NN). Among the three architectures, 1D-Convolutional Neural Network(1D-CNN) performed best with a test accuracy of 97.9%.