MULTI-LEVEL STACKING ENSEMBLE LEARNING FOR ENHANCED ECG-BASED ARRHYTHMIA DETECTION

Abstract

Cardiovascular diseases (CVDs), characterized by abnormal heart function, are a leading cause of mortality worldwide, emphasizing the critical need for early detection and intervention. Electrocardiogram (ECG) monitoring has emerged as a vital tool for heart rhythm assessment in CVD prevention, with recent Machine Learning applications significantly improving diagnostic accuracy. This paper presents an innovative multi-level stacking ensemble learning approach that integrates predictions from diverse base models, including Random Forest, XGBoost, Support Vector Machine, K-Nearest Neighbor, AdaBoost, and Decision Tree algorithms, applied to 12-lead ECG data preprocessed using Neurokits. The base model outputs are synthesized through a Logistic Regression meta-model to enhance overall performance. Our methodology demonstrates exceptional effectiveness across multiple evaluation metrics, including accuracy, F1-score, precision, recall, and specificity, achieving over 97% accuracy in classifying various arrhythmia types. This research underscores the significant potential of ensemble learning methods in cardiac diagnostics, offering robust and comprehensive predictions for complex clinical scenarios.