Detection of Ventricular Arrhythmias using HRV Analysis and Quadratic Features

Background: Ventricular Arrhythmias, one of the fatal heart diseases, requires timely recognition. The nonlinear and random nature of heart rate makes the diagnosis challenging. Introduction: The research work in this paper is divided into three phases. In the first phase, classification of some of the ventricular arrhythmias is done in four classes as Premature Ventricular Contraction (PVC), Left Bundle Branch Block (LBBB) and Right Bundle Branch Block (RBBB) with some Normal (N) samples and the analysis of classifying algorithms to improve the classifiers accuracy. A Support Vector Machine (SVM), Random Forest, Artificial Neural Network (ANN), and K Nearest Neighbor (KNN) algorithms were used to train and test the classifier, with the help of online available MIT-BIH Arrhythmia Database. Then, in the second phase, the variance analysis of the data is carried out using Principle Component Analysis (PCA) to improve the classifier performance. In the last phase, the whole process is repeated after including Quadratic features with the best performing classifier only. Methods: Signal processing, generation of Heart Rate Variability (HRV) signals from the available Electrocardiogram (ECG) signals and training, testing of ANN classifier was done in MATLAB environment, and the training and testing of SVM, and Random Forest classifier was done in R project software. Results: Random Forest shows the best result among all classifiers with 86.11% accuracy, 87.1% after applying PCA with top 16 features, and 91.4% after including quadratic features with top 28 features. Conclusion: The present study envisages helping ECG and HRV data analyses while selecting the AI techniques for classification purposes according to data.