Akademik Veri Yönetim Sistemi
5th International Conference on Informatics and Software Engineering, IISEC 2026, Ankara, Türkiye, 5 - 06 Şubat 2026, ss.462-467, (Tam Metin Bildiri)
Accurate and reliable classification of cardiac arrhythmias from electrocardiogram (ECG) signals is essential for early diagnosis and continuous cardiac monitoring. However, ECG-based arrhythmia classification remains challenging due to severe class imbalance, inter-patient variability, and the limited interpretability of deep learning models. In this study, a robust and explainable deep learning framework for multi-class ECG heartbeat classification is proposed and evaluated on the MIT- BIH Arrhythmia Database. The framework operates directly on 1D ECG signals and integrates systematic preprocessing, stratified data splitting, and data augmentation to address class imbalance. Multiple deep learning architectures, including BiLSTM, ResNet-1D, Inception-1D, and CNN with attention, are trained under a unified experimental setup. Experimental results demonstrate that the ResNet-1D model achieves the best overall performance, attaining an accuracy of 99.18% and an F1-score of 94.60%. To enhance transparency and clinical interpretability, Grad-CAM and SHAP-based explainability techniques are incorporated to analyze model decision behavior. The XAI results consistently indicate that the proposed models focus on physiologically meaningful ECG segments, particularly the QRS complex. Overall, the proposed framework demonstrates that high classification performance and interpretability can be jointly achieved for multi-class ECG heartbeat classification.