ECG Heartbeat Classification: A Comparative Analysis of ML Ensemble Techniques
Chennaiah Kate, Deepika Agrawal, Tirath Prasad Sahu
Abstract
Electrocardiogram (ECG) signals play a vital role in diagnosing and analysing cardiovascular conditions. Accurately pinpointing and categorising heartbeats within ECG data is essential for the precise identification of arrhythmias and other heart abnormalities. This study conducts a comparative assessment of various ensemble machine learning techniques, including Random Forest, Extra Trees, XGBoost, Gradient Boosting, Voting, and Stacking, for classifying ECG heartbeats. This research examines how dimensionality reduction and data balancing methods affect classification accuracy. The oversampling method Adaptive Synthetic(ADASYN) was utilised for class imbalance, and Principal Component Analysis (PCA) was employed to reduce feature dimensions. The model's effectiveness was assessed using the MIT-BIH Arrhythmia dataset, with evaluation metrics including accuracy, confusion matrix, precision, recall, F1-Score classification report, and AUC-ROC. The Extra Trees achieved the highest performance with a test accuracy of 99.78%, a recall of 99.78%, an F1-score of 99.78% and an ROC-AUC score of 100%. The other models, Random Forest, CatBoost, and XgBoost, also produced comparably high results with accuracies of 99.70%, 97.20%, and 99.30%, respectively. These results confirm the strength and generalizability of ensemble methods, making them promising candidates for ECG-based diagnostic systems.