Multiple-Time-Scale Data Augmentation and Application in AF detection

Abstract

Data augmentation is an effective method for fully training a model with limited data in machine-learning applications. In this study, a novel multiple-time-scale data augmentation method was developed that first extracts some features of samples on multiple time scales and then takes the stationary features with various time scales of a single sample as the features of different samples to achieve data augmentation. Finally, the method was applied to the detection of atrial fibrillation (AF). The long-term ECG signals from AF patients and control group in PhysioBank were used as the training set, and then the long-term ECG signals collected in dynamic ECG room of Shandong Provincial Hospital were used as the test set, finally classified by support vector machine (SVM). The classification accuracy of training set and test set were 98.06% and 93.33%, respectively. Which proved the effectiveness of the proposed data augmentation method in AF detection and its strong generalization ability.