Classification ECG Signals Base on kNearest Neighbors (k-NN) Algorithm

Abstract

Abnormal cardiac rhythm known as atrial fibrillation (AF) is marked by an atria's fast and erratic pulse. It often begins in short periods of abnormal beating which becomes longer and may be constant over time. Usually it presents no symptoms and a typical ECG affected by Atrial Fibrillation does not present any P wave and shows an irregular ventricular rate. In this study, the k-Nearest Neighbors (K-NN) algorithm has been used to classifier 5000 samples of cardiac signals. After preprocessing the data, it was split into the three classes represented, namely: Normal (N), AF, and Noisy Rhythm (NR). In a ratio of 1:1, the data were split into two groups: training dataset and test dataset, to perform the classification. It was obtained from the dataset, the highest sensitivity recorded for N cases is 92% and the highest specificity recorded for AF is 99%. The classification accuracy obtained is 90% and the value for area under the curve (AUC) is 0.94