Blending Ensemble Learning for Enhanced Arrhythmia Classification Utilizing 12-Lead ECGs

The rise in heart-related diseases has driven the development of advanced techniques for identifying irregular heart conditions, and with the progress in artificial intelligence and signal processing, automated arrhythmia classification using machine learning and electrocardiograms (ECG) has become increasingly effective and widely utilized by healthcare professionals. This paper presents an efficient machine learning solution for robust arrhythmia classification using 12-lead electrocardiograms (ECGs) by leveraging blending ensemble learning. Our developed blending model utilizing six base models-Adaptive boosting, Extreme gradient boosting, Decision trees, K-nearest neighbors, Random forest, and Support vector machine-with a Logistic regression meta-model, demonstrates enhanced efficiency in classifying arrhythmias based on 12-lead ECGs. This approach not only exploits the unique strengths of each base model but also captures diverse predictive patterns, which are combined by the LR meta-model into a refined and cohesive output. The use of LR as the meta-model enhances interpretability and generalization, reducing the risk of overfitting and optimizing overall performance. Experimental results demonstrate that the proposed blending ensemble outperforms conventional notable works in terms of accuracy and offers a robust and effective solution for accurate arrhythmia classification, i.e. up to 97.2% accuracy, supporting clinical decision-making.