Feature Extraction Methods for Predicting the Prevalence of Heart Disease

Abstract

AbstractThis paper presents an automatic classification technique for the detection of cardiac arrhythmias from ECG signals. With cardiac arrhythmias being one of the leading causes of death in the world, accurate and early detection of beat abnormalities can significantly reduce mortality rates. ECG signals are vastly used by physicians for diagnosing heart problems and abnormalities as a result of its simplicity and non-invasive nature. The aim of this study is to determine the most accurate combination of feature extraction methods and SVM (Support Vector Machine) kernel classifier that will produce the best results on ECG signals obtained from the MIT-BIH Arrhythmia Database. SVM classifiers with four different kernels (linear, polynomial, radial basis, and sigmoid) were used to classify different features extracted from the four feature selection methods; Random Forests, XGBoost, Principal Component Analysis, and Convolutional Neural Networks. The CNN-SVM classifier produced the best results overall, with the polynomial kernel achieving the maximum accuracy of 99.2%, the best sensitivity 92.40% from the radial basis kernel, and best specificity of 98.92% from the linear kernel. The high classification accuracy obtained is comparable to or even better than other approaches in literature.KeywordsFeature extractionRandom Forest (RF)Boosted Trees (BT)Principal Component Analysis (PCA) and Convolutional Neural Networks (CNN)Support Vector Machines (SVM)Heart diseaseArrhythmiasECGClassification