Data Augmentation on Marginal samples and distribution-based selection to improve ECG signal classification with deep learning

Articles in Press

Document Type : Original Article

Authors

1 Faculty of Computer Engineering Shahid Rajaee Teacher Training University, Iran

2 Department of Computer Engineering, Islamic Azad University, Torbat Heydariyeh Branch Torbat Heydariyeh, Iran

Abstract

Heart diseases are one of the most common types of diseases that create high mortality rates. Arrhythmias are abnormal beats that make the heart work too fast (tachycardia) or too slowly (bradycardia) and pump ineffectively. The existence of unbalanced datasets makes the detection of some types of arrhythmias a challenge. The aim of this research is to provide a solution based on data augmentation for deep networks in order to classify arrhythmias that have small sample sizes in their dataset. As such, in order to boost deep learning in the classification of cardiovascular signals, an innovative method in data augmentation is presented in which a mechanism is considered to select the most important initial samples to perform common data augmentation of time series on them and to produce artificial samples. Next, this mechanism is applied again on the generated artificial samples and more valuable samples are selected based on proper distribution. Following, these valuable samples are used to train deep convolutional neural networks. The obtained results showed that the presented model improved the classification results in 17 classes of MIT-BIH data and especially the arrhythmias with small samples. In addition, it was able to achieve 96.92% accuracy in the classification of 17 classes of MIT-BIH dataset that had an imbalance rate of 28.3. The presented method can be used in data augmentation of the other periodic time series data.

Keywords

Main Subjects

Advanced Signal Processing

Articles in Press, Accepted Manuscript
Available Online from 16 August 2025

Files

Share

How to cite

Statistics