Abstract
Early diagnosis of cardiovascular diseases (CVDs) is essential for improving patient outcomes. As a primary diagnostic modality, electrocardiogram (ECG) signals pose challenges for automatic classification due to their complex temporal and morphological characteristics. This study proposes a CNN-CBAM-GRU model that integrates Convolutional Neural Networks (CNN), the Convolutional Block Attention Module (CBAM), and Gated Recurrent Units (GRU) to enhance both spatial feature representation and temporal sequence modeling. The model is evaluated on two public ECG datasets-MIT-BIH and PTB-XL-under five-class classification settings. Unlike many existing approaches that report only a limited set of metrics, this study conducts a comprehensive evaluation across multiple performance indicators, including accuracy, precision, recall, sensitivity, and F1-score, providing a more complete view of classification effectiveness. Experimental results demonstrate that the proposed model achieves a strong balance between predictive performance and computational efficiency. Specifically, it achieves 98.17% accuracy and 98.91% F1-score on MIT-BIH, and 99.21% accuracy and 99.47% F1-score on PTB-XL, with a compact parameter size of 2.45 million. These findings validate the proposed model as a practical and robust solution for intelligent ECG classification and automated cardiovascular disease diagnosis.