Evaluation study on Cardiac Abnormalities using Ensemble Classifier and Convolutional Neural Network

A preliminary screening technique for identifying heart problems is the Electrocardiogram (ECG). With computer-assisted diagnostics, anomalies are detected earlier and prediction accuracy is increased. In this work, two models are developed, one using machine learning and the other using deep learning to classify cardiac abnormalities. The MIT-BIH Arrhythmia database is used to create the model. It comprises the five categories of abnormal beats which include Left Bundle and Right Bundle Branch Block, Ventricular Premature Contraction, and Atrial Premature Contraction. For the machine learning model, seven spectral and four statistical components are retrieved from the decomposed ECG signal. The voting ensemble classifier is developed using Random Forest, Support Vector Machine, XGBoost, and Light GBM, which produced an accuracy of 98.52%, recall of 98.42%, precision of 98.51%, and F1 score of 98.34% respectively. Also, a Convolutional Neural Network is implemented using signals from different combinations of intrinsic mode functions (IMFs) using the Empirical Mode Decomposition (EMD) and the accuracy obtained is 99.03%. The proposed method can help clinicians to predict the abnormalities at the earliest using ECG signals.