Multi-Scale Convolutional Neural Network with LSTM for Accurate ECG Beat Classification
International Journal of Research & Technology · 03.07.2026 · 14 (3) · 81-99
DOI: 10.64882/ijrt.v14.i3.1583 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Research & Technology, online erschienen am 03.07.2026. Verfasst von Dr. Hemant Amhia und 1 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Electrocardiogram (ECG) signal classification plays a vital role in the early diagnosis of cardiac arrhythmias by enabling the automatic identification of abnormal heartbeats. However, the nonlinear nature of ECG signals and the presence of imbalanced heartbeat classes make accurate classification a challenging task. This paper proposes a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) framework for automated ECG heartbeat classification using the MIT-BIH Arrhythmia Database. Initially, the ECG signals undergo preprocessing, including data validation, label conversion, and reshaping into one-dimensional sequences. To address the class imbalance problem, Random Over Sampling (ROS) is applied to generate a balanced training dataset. The proposed architecture employs three successive one-dimensional convolutional layers with batch normalization and max-pooling operations to extract discriminative morphological features from ECG waveforms. Subsequently, two stacked LSTM layers are utilized to capture the temporal dependencies and sequential characteristics of heartbeat signals. The extracted deep features are then processed through fully connected dense layers, while a Softmax classifier performs five-class heartbeat classification. The network is trained using the Adam optimizer with categorical cross-entropy loss, together with early stopping and adaptive learning-rate reduction to enhance convergence and reduce overfitting. Performance evaluation is carried out using the independent MIT-BIH test dataset based on accuracy, precision, recall, F1-score, and confusion matrix analysis. Experimental results demonstrate that the proposed CNN–LSTM model achieves an overall classification accuracy of approximately 98%, outperforming conventional machine learning and standalone deep learning approaches in terms of robustness and generalization. The proposed framework provides an efficient and reliable computer-aided diagnosis system for accurate arrhythmia detection and has significant potential for real-time clinical monitoring and intelligent healthcare applications.
Zitieren
APA
Amhia, D. H., & Mishra, R. (2026). Multi-Scale Convolutional Neural Network with LSTM for Accurate ECG Beat Classification. International Journal of Research & Technology, 14(3), 81-99. https://doi.org/10.64882/ijrt.v14.i3.1583
Vancouver
Amhia DH, Mishra R. Multi-Scale Convolutional Neural Network with LSTM for Accurate ECG Beat Classification. International Journal of Research & Technology. 2026;14(3):81-99. doi:10.64882/ijrt.v14.i3.1583
BibTeX
@article{Amhia2026,
author = {Amhia, Dr. Hemant and Mishra, Rahul},
title = {Multi-Scale Convolutional Neural Network with LSTM for Accurate ECG Beat Classification},
journal = {International Journal of Research & Technology},
year = {2026},
volume = {14},
number = {3},
pages = {81--99},
doi = {10.64882/ijrt.v14.i3.1583}
}
EUVIMED