Comparative Analysis of CNN and CNN+Transformer Based Arrhythmia Classification on One-Dimensional ECG Signals Tek Boyutlu EKG Sinyallerinde CNN ve CNN+Transformer Tabanli Aritmi Siniflandirmasinin Karsilastirmali Analizi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636605
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: arrhythmia classification, CNN, deep learning, ECG, Transformer
- İstanbul Yeni Yüzyıl Üniversitesi Adresli: Evet
Özet
Early and accurate detection of cardiac arrhythmias is crucial for preventing life-threatening cardiovascular events. In this study, five-class arrhythmia classification was performed using single-beat ECG segments derived from the MIT-BIH arrhythmia database. The main objective was to compare a pure convolutional approach, namely CNN, with a hybrid CNN+Transformer architecture on one-dimensional ECG signals. Each ECG beat was represented by 256 samples and normalized using sample-wise z-score normalization. The evaluation was carried out using stratified 5-fold cross-validation, where 10% of the training data in each fold was reserved for validation. To reduce the effect of class imbalance, class-weighted loss was employed during training. Experimental results were evaluated using accuracy, macro F1-score, weighted F1-score, balanced accuracy, and class-wise precision-recall-F1 metrics. The findings indicate that the CNN model provides a strong balance between classification performance and computational efficiency, while the CNN+Transformer model offers competitive results on certain classes along with potential interpretability advantages.