Automatic arrhythmia identification based on electrocardiogram data using hybrid of Support Vector Machine and Genetic Algorithm

Abstract

Electrocardiogram (ECG) recordings provide insights on a person's cardiac activity, and can be used to identify heartbeat abnormalities, or arrhythmia, they might suffer. Automatic ECG interpretation can be achieved via machine learning techniques to aid physicians. This research aims to model a ECG classifier based on Support Vector Machine (SVM) and Genetic Algorithm (GA) to classify a normal beat and three types of arrhythmia. SVM with radial basis function (RBF) kernel were used because of its superiority in handling the large number of numerical features generated from the ECG. GA was used to enhance the SVM by performing feature selection to generate dataset with optimal number of features, limiting possibilities of feature redundancy. ECG dataset used for model training and testing were obtained from the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) arrhythmia database. The GA-SVM classifier then was compared to SVM-only classifier to ensure that GA is indeed able to improve the capabilities of SVM. Results show that in classifying six-seconds ECG recording with 120 training data and 20 testing data, GA-SVM yielded better average accuracy of 82.5%, compared to 47% yielded by SVM-only classifier.