ECG Classification Using Machine Learning Classifiers with Optimal Feature Selection Methods

Abstract

Machine Learning (ML) is a booming technology well-suited for data analysis in all fields, especially healthcare. Cardiovascular Disease (CVD) is one of the deadliest diseases that has a high mortality rate. Irregular beats of the heart are known as Cardiac Arrhythmia (CA), and its diagnosis is a challenging task for cardiologists. Feature selection is an essential and inevitable process that identifies the causing factors of arrhythmia. The proposed model proves that arrhythmia prediction is possible with a limited set of features obtained from Electrocardiogram (ECG) signals. Firstly, the standard UCI Machine Learning Arrhythmia dataset is subject to preprocessing and normalization after applying optimal feature selection techniques such as Spearman Ranking Coefficient and Mutual Information. Eventually, the result was a reduced dataset fed as the input for eight robust Machine Learning classifiers such as Decision Tree (DT), Adaboost (AB), K-Nearest Neighbour (K-NN), Naive Bayes (NB), Random Forest (RF), Gradient Boosting (GB), Support Vector Machine (SVM), and Multilayer Perceptron (MLP). Analysis of the results presents the highest accuracy of 91.3% with the Spearman feature selection technique and Random Forest Classifier. Furthermore, the proposed work shows better achievement when compared with state-of-the-art methods.