Comparison of 1D-Convolution Neural Network and Hilbert Huang Transform-Based Features for ECG Signal Classification

The electrocardiogram (ECG) signal is one of the primary biological signals used in the diagnosis of cardiac arrhythmias. However, conventional analysis and classification methods are often limited by the nonlinear and non-stationary nature of ECG signals. This research aims to evaluate the performance of a one-dimensional convolutional neural network (1D-CNN) and hybrid models that integrate Hilbert–Huang Transform (HHT)-based features with 1D-CNN for ECG signal classification. Five standard diagnostic categories (N, S, V, Q, and F) were considered. A comparative framework was developed, including a standalone 1D-CNN model and three hybrid models based on different HHT-derived features. Model performance was assessed using confusion matrices, ROC-AUC curves, and key statistical metrics, including accuracy, precision, recall, and F1-score. The results indicate that the standalone 1D-CNN model achieved the highest classification performance when applied to raw ECG signals. Among the hybrid models, the Mean_E + 1D-CNN model demonstrated superior performance compared to the other feature-based models. These findings suggest that while HHT features provide valuable time–frequency interpretability, direct learning from raw signals using deep learning models remains more effective for ECG classification.