A Selective Ensemble Learning Framework for ECG-Based Heartbeat Classification with Imbalanced Data

Abstract

ECG-based heartbeat classification is often accompanied with difficult feature extraction and imbalanced sampling data. In order to alleviate the bias in performance caused by imbalanced data, a Selective Ensemble Learning Framework based on sample Distribution and classifier Diversity (SELFrame-DD) is proposed for ECG-based heartbeat classification. In SELFrame-DD, an improved SMOTE algorithm is proposed to generate training sets by using a sample-distribution based resampling strategy, and the selective ensemble depends on the diversity of classifiers and the prediction accuracy of classifiers for minority classes. Besides, a multimodal ECG feature extraction is employed based on wavelet packet decomposition and 1-D convolutional neural network. Experimental studies on MIT-BIH arrhythmia database show that the proposed algorithm can achieve a high classification accuracy for imbalanced multi-category classification.