Key result
Feature-based machine learning classifies ECG arrhythmias with ~89% accuracy across patient-independent datasets.
Why the study?
Many deep-learning approaches for automated ECG heartbeat classification require substantial computational resources and lack interpretability.
Does an interpretable feature-based machine-learning framework using Random Forest, XGBoost, and CatBoost accurately classify ECG heartbeats under patient-independent evaluation in the MIT-BIH Arrhythmia Database?
Does an interpretable feature-based machine-learning framework using Random Forest, XGBoost, and CatBoost accurately classify ECG heartbeats under patient-independent evaluation in the MIT-BIH Arrhythmia Database?
Interpretable feature-based machine learning models using 20 morphological and temporal ECG features can achieve high accuracy (up to 89.34%) and strong discrimination for ventricular ectopic and fusion beats under patient-independent evaluation, though supraventricular ectopic beats remain challenging.
*Abstract*—Automated Electrocardiogram (ECG) heartbeat classification plays an important role in arrhythmia diagnosis and clinical decision-support systems. Although deep-learning approaches have recently achieved remarkable performance, many require substantial computational resources and lack interpretability. This study proposes an interpretable feature-based machine-learning framework for automated heartbeat classification using the MIT-BIH Arrhythmia Database. Twenty physiologically meaningful morphological and temporal features were extracted from each heartbeat. Random Forest, XGBoost, and CatBoost, were evaluated under a rigorous patient-independent inter-record validation protocol. Random Forest achieved the highest overall accuracy (89.34%), whereas CatBoost obtained the highest weighted precision (90.90%), weighted F1-score (89.28%), and macro F1-score (40.77%). ROC analysis demonstrated strong discrimination for ventricular ectopic beats (AUC = 0.986) and fusion beats (AUC = 0.937), whereas supraventricular ectopic beats remained challenging (AUC = 0.653) because of their morphological similarity to normal beats and severe class imbalance. Feature-importance analysis identified the previous RR interval and waveform statistical descriptors among the most influential predictors. These findings demonstrate that interpretable feature-based machine learning can provide computationally efficient ECG classification under patient-independent evaluation, although external validation and real-time implementation studies are required before clinical deployment.
No takes yet. Share an insight, caveat, or question.
Leila Sahraoui (2026) studied Cardiac arrhythmia. Interpretable feature-based machine-learning framework (Random Forest, XGBoost, and CatBoost using 20 morphological and temporal ECG features) vs. Comparison across Random Forest, XGBoost, and CatBoost models was evaluated on Automated ECG heartbeat arrhythmia classification accuracy, precision, F1-score, and ROC AUC (AUC 0.986 for ventricular ectopic beats; AUC 0.937 for fusion beats; AUC 0.653 for supraventricular ectopic beats). Feature-based machine learning achieved 89.34% accuracy (Random Forest) and 90.90% weighted precision (CatBoost) for patient-independent ECG arrhythmia classification.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:
More on Artificial intelligence in cardiologyMore from International Journal of Electrical and Electronic Engineering & Telecommunications