Development of a Variety of Fast Machine Learning Model for ECG-based Arrhythmia Classifier

Although deep learning has been proving its capability in various fields, training and testing by learning a large amount of data and deep neural networks remain time consuming. To address this issue, a high-performance GPU and CPU, SSD storage, and a large amount of RAM is required, which is expensive. We propose a new classifier algorithm by feature point extraction that can be trained and tested quickly. The performance of the proposed algorithm was verified by classifying heart diseases by applying the MIT-BIH arrhythmia data set. First, the noise was removed by Wavelet transform, and feature points were extracted using root mean square (RMS), crest factor, margin factor, form factor, kurtosis, and pulse factor. Then, the performance was compared using various classification algorithms. The two feature extraction methods are compared to evaluate the accuracy of each algorithm, the execution time of the model during training, and the memory usage. Our proposed algorithm is applied to various health care systems such as heart disease and depression, and it is predicted that it will be able to help users toward health care at low cost.