Design and implementation of ECG classification based on heart rate variability analysis

Heart rate variability (HRV) analysis is a non-invasive method derived from electrocardiogram (ECG) data that reflects the activity of both the sympathetic and vagus nerves. To conduct HRV analysis, this study first presents an R-peak detection algorithm which consists of digital filtering, Shannon energy envelope and Hilbert transform. Next, heart rate variability analysis including time domain analysis, frequency domain analysis and nonlinear analysis is calculated. After that, the features are fed into a LightGBM classifier for multi-class classification. The MIT-BIH arrhythmia database is used for evaluation. The R-peak detection algorithm achieves average detection accuracy of 99.78%, precision of 99.93%, recall of 99.84%. Significance tests demonstrate that heart rate variability analysis indicates a significant difference between the normal group and the arrhythmia group. The multi-class arrhythmia classification achieves average precision of 98.18%, recall of 98.17% and accuracy of 96.43%.