Machine Learning Techniques with Low-Dimensional Feature Extraction for Improving the Generalizability of Cardiac Arrhythmia

摘要

Automatic heartbeat classification is an important stage in identifying cardiac arrhythmia. Several machine learning (ML) techniques have been proposed to perform this, but they produce an accuracy result of below 99%. In this study, a deep neural network (DNN) structure is applied to improve ML performance. The feature selection method is based on the combination of discrete wavelet transform (DWT) and principal component analysis (PCA). To avoid computational complexity, the components of PCA are derived by low-dimensional DWT coefficients. The results show that the proposed ML model achieves good performance, producing 99.76% accuracy, 91.80% sensitivity, 99.78% specificity, 93.02% precision, and 92.31% F1-score. To benchmark the proposed model, the support vector machine (SVM) and random forest (RF) techniques are used as the baseline models. The DNNs are 2.3% more sensitive than SVM, while the RF fails to classify the ECG heartbeat. Four datasets are used to analyze the robustness and generalization performance of the proposed model: MIT-BIH, SVDB, MITDB, and IncartDB. All testing results produce satisfying performance. The proposed ML model offers a potential solution to improve the generalizability of a DNN-based model in different cardiac datasets for classifying tasks.