Comparison of Several Wavelet Transform Techniques for Inferring ECG Signal Classification

Cardiovascular diseases (CDVs) remain a leading cause of death worldwide. Thus, the disease needs to be detected as early as possible. One of the solutions is using an automated ECG Signal prediction using machine learning or deep learning. Among the many techniques in automated ECG signal analysis, wavelet transform has been highlighted due to its effectiveness in denoising and feature extraction. Then, a Convolutional Neural Network (CNN) is performed to predict the heartbeat. In the group of many techniques in wavelet transform, it is necessary to find out which is better in heartbeat prediction not only to be trained and tested with a secondary dataset but also with a primary or real dataset of ECG signal. In this paper, we compare the wavelet transform techniques, such as Continuous Wavelet Transform (CWT), Discrete Wavelet Transform (DWT), and Stationary Wavelet Transform (SWT), and also do inferencing. Those techniques were trained and tested by MIT-BIH arrhythmia database. As a result, the highest test accuracy is achieved by CWT with 0.9908 accuracy. Then, by testing it with a 30-minute recording of the ECG signal, the shortest time in inferencing is achieved by DWT in 12.92 seconds. It also used the lowest memory with 109.98 MB and SWT achieved the lowest CPU usage in inferencing with 26.34%.