A comparative study of multivariate approach with neural networks and support vector machines for arrhythmia classification

This paper presents a comparative study of multivariate approach i.e. principal component analysis (PCA) for ECG signal analysis with support vector machine (SVM) and back propagation neural network for classification. Here, the combination of different sets of feature extraction and classification algorithms are analyzed and compared with each other to yield the best performance in terms of accuracy and other performance metrics. The experiment is performed to classify six classes of ECG beats and evaluated using the MIT-BIH database. The results show that the kernel based PCA with support vector machine performs better with an average overall accuracy, sensitivity, specificity and positive predictivity of 98.96%, 98.90%, 99.79% and 98.98% respectively.