Key result
Extra Trees tops 21 machine-learning models for single-lead ECG arrhythmia detection with ~98% accuracy.
Why the study?
Automatic classification of cardiac abnormalities from single-channel ECG signals on connected monitoring devices requires comparing machine-learning classifiers across both predictive performance and computational requirements.
Which machine-learning classifier provides the optimal balance of predictive performance, statistical stability, and computational efficiency for single-lead ECG arrhythmia classification in connected cardiac monitoring devices?
Which machine-learning classifier provides the optimal balance of predictive performance, statistical stability, and computational efficiency for single-lead ECG arrhythmia classification in connected cardiac monitoring devices?
Answers are grounded in this paper and cite their sources.
Extra Trees and LightGBM offer high accuracy, statistical stability, and favorable computational efficiency for automated single-lead ECG arrhythmia classification in resource-constrained IoT monitoring devices.
This study compares twenty-one machine-learning classifiers for the automatic classification of cardiac abnormalities from single-channel ECG signals, with particular attention to predictive performance and computational requirements relevant to connected cardiac monitoring devices. Five heartbeat classes are considered: normal beat (NOR), atrial premature contraction (APC), left bundle branch block (LBBB), right bundle branch block (RBBB), and ventricular premature contraction (VPC). ECG signals were extracted from the MIT-BIH Arrhythmia Database and segmented into one-second windows centered on annotated beat positions. The preprocessing pipeline included wavelet-based denoising, normalization, and class balancing using a hybrid RandomUnderSampler–SMOTE strategy. All twenty-one classifiers were evaluated using stratified five-fold cross-validation on the training partition, while an independent held-out test set was preserved for complementary evaluation. Performance was assessed using accuracy, macro precision, macro recall, macro F1-score, macro-AUROC, training time, and inference latency. Extra Trees achieved the highest mean cross-validation accuracy (98.12±0.16%) and macro F1-score (95.05±0.52%), together with the best average rank across folds. Random Forest, XGBoost, LightGBM, and SVM also achieved strong predictive performance. The Friedman test revealed significant global differences among the evaluated classifiers (χ2=98.0468, p=2.8124×10−12), whereas the subsequent Nemenyi analysis indicated that the leading classifiers were statistically comparable at the 0.05 significance level. Considering predictive performance, stability across cross-validation folds, and computational cost, Extra Trees and LightGBM were retained as promising candidates for further investigation in resource-constrained ECG monitoring applications. These results emphasize that classifier selection for future embedded biomedical systems should consider not only predictive performance, but also statistical stability and computational requirements.
No takes yet. Share an insight, caveat, or question.
Abdel-Alli et al. (2026) studied Cardiac arrhythmias (NOR, APC, LBBB, RBBB, VPC). Extra Trees machine-learning classifier vs. Twenty other machine-learning classifiers (including Random Forest, XGBoost, LightGBM, and SVM) was evaluated on Mean cross-validation classification accuracy and macro F1-score (p=2.8124x10-12). Extra Trees achieved the highest mean cross-validation accuracy and macro F1-score among 21 classifiers for single-lead ECG arrhythmia classification.
Synapse has enriched 10 related papers. Consider them for comparative context: