An Explainable Attention-Based Hybrid Learning Framework for the Multi-Class Classification of ECG-Based Arrhythmia and Cardiovascular Diseases

Electrocardiogram (ECG) analysis plays a vital role in the early diagnosis of cardiac arrhythmias and cardiovascular diseases (CVDs), which remain the leading cause of global mortality. Recent advances in machine learning (ML) and deep learning (DL) have significantly improved automated ECG classification accuracy. However, most existing DL models function as black-box systems, limiting their adoption in clinical environments due to the lack of interpretability and transparency. This paper proposes an explainable attention-based hybrid deep learning and machine learning ensemble framework for ECG-based multi-class arrhythmia and disease classification. A CNN–BiLSTM architecture enhanced with an attention mechanism is employed for spatial–temporal feature extraction, while AdaBoost and Support Vector Machine (SVM) classifiers are integrated using a stacking ensemble strategy. Experiments conducted on the MIT-BIH Arrhythmia and PTB Diagnostic ECG datasets demonstrate an overall classification accuracy of 99.53%, outperforming existing ML and DL approaches. Explainable artificial intelligence (XAI) techniques such as Grad-CAM and SHAP are incorporated to provide clinically interpretable predictions, thereby improving trust and reliability in automated ECG diagnosis.