A Hybrid CNN–BiLSTM–LSTM–Attention Framework for Multichannel ECG-Based Cardiac Arrhythmia Classification

This paper proposes a deep learning–based approach for cardiac arrhythmia classification from multichannel electrocardiogram (ECG) signals using a hybrid architecture that integrates Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Long Short-Term Memory (LSTM), and a Multi-Head Self-Attention mechanism. The input data consist of four-channel ECG signals stored in CSV format and continuously recorded over time. Prior to model training, the ECG signals undergo a comprehensive preprocessing pipeline, including amplitude clipping within ±150 µV to remove impulsive artifacts, a 50 Hz notch filter for power-line interference suppression, a 0.5–45 Hz bandpass filter to preserve the physiological frequency components of ECG signals, and segment-wise normalization. After preprocessing, the signals are segmented into 4-second windows with 50% overlap, generating input samples of size (1024, 4). The segmented ECG windows are first processed by a one-dimensional CNN to extract local morphological features. Subsequently, BiLSTM and LSTM layers are employed to learn temporal dependencies within the signal sequences. The Multi-Head Self-Attention mechanism enables the model to focus on the most informative temporal regions of each ECG segment. Finally, a classification module consisting of Global Average Pooling, Dense, and Softmax layers performs heartbeat classification. Experimental results demonstrate that the proposed model achieves high classification performance in terms of Precision, Recall, and F1-score on the test dataset. These findings indicate that the combination of an effective ECG preprocessing strategy and a CNN–BiLSTM–LSTM–Attention architecture provides a promising solution for multichannel ECG-based arrhythmia classification..