Evaluation of ECG based Recognition of Cardiac Abnormalities using Machine Learning and Deep Learning

Around the world, the most common cause of death is due to heart disease. To reduce the risk of death, it is critical to analyze and predict heart diseases. The proposed approach introduces a novel technique to detect anomalies in Electrocardiogram signals and classify cardiac conditions (17 – classes) from 1000 fragments of ECG signals from 45 patients in the MIT-BIH Arrhythmia database. The proposed approach utilizes two methods, one approach based on conventional Machine learning algorithm i.e., SVM (Support Vector Machine) and while the deep learning method uses CNN (Convolutional Neural Network) based architecture (ALEXNET). Spectral power density was detected using Welch's process, and Discrete Fourier Transform. The data was standardized and scaled to a standard deviation of unity. On the test set, CNN classification accuracy was reported to be 87-90%, while SVM classification accuracy was recorded to be 70-76%. According to the literature this is the simplest approach with state-of-the-art deep learning method to achieve accuracy up to 90%. The deep learning technique improves precision and can be used in clinical settings.