Single and Multi-Lead ECG Heartbeat Classification Using XGBoost on Pre-trained CNN Feature Layer

Abstract

Electrocardiogram (ECG) is a common tool used in technology-assisted systems to help spot heart rhythm problems. Doctors rely on it to identify and understand any irregularities in the heart's activity. Therefore, making the process of analyzing ECG heartbeats automatic is crucial for diagnosing heart conditions effectively. This proposed work introduces a groundbreaking approach for analyzing electrocardiogram (ECG) signals, essential for diagnosing heart conditions. It uniquely combines XGBoost, a potent predictive model, with the capabilities of pretrained convolutional neural networks (CNNs) for detecting features in ECG data. Applied to both single and multi-lead ECG datasets, specifically MIT-BIH and PTB-XL, this approach offers a straightforward yet highly effective way to diagnose cardio vascular diseases. Compared with traditional models like Decision Trees, Random Forest, and others, the proposed method significantly outperforms, achieving accuracies of 95.8% for single-lead and 97% for multi-lead signals. This innovation excels in precision and reduces the complexity and computational needs, showcasing a significant leap forward in efficient and accurate ECG classification.