Classification of the Heartbeats in Electrocardiograms with K-Nearest Neighbors Algorithm, Random Forests, and Support Vector Machines - A Pilot Study

Abstract

Electrocardiography (ECG) is vital for diagnosing and monitoring cardiovascular diseases. Automatic interpretation of ECG signals, particularly classifying heartbeats, is essential in diagnosing these conditions. We compared the performance of three classifiers (K-Nearest Neighbors (KNN), Random Forests (RF), and Support Vector Machine (SVM)) using heartbeat and ECG signal features. We trained the classifiers on 173 heartbeats from 11 recordings in the MIT-BIH Arrhythmia (MITDB) Database. To address imbalance, we used SMOTE (Synthetic Minority Over-sampling Technique) to increase the number of abnormal heartbeats. We divided the data into 70% for training set and 30% for testing set, evaluating performance through 10-fold cross-validation. Results demonstrated state-of-the-art performance for KNN, RF, and SVM. KNN achieved an overall accuracy, positive predictive value, and sensitivity of 0.977, 0.978, and 0.978, respectively. RF achieved an overall accuracy, precision, and sensitivity of 0.971, 0.972, and 0.971, respectively. SVM achieved a recall of 0.988, 0.989, and 0.988. SVM exhibited the highest overall performance in heartbeat classification (0.99), followed by KNN (0.98) and RF (0.97).