Arrhythmia Classification Using Continual Normalization-Based Gated Recurrent Unit on the Chapman-Shaoxing Dataset

DOI

10.2991/978-94-6239-828-3_26How to use a DOI?

Keywords

Gated Recurrent Unit; Continual Normalization; Autoencoder; Chapman-Shaoxing; ECG

Abstract

Cardiac arrhythmias are a leading cause of global mortality, and ECG is the primary non-invasive detection tool, yet manual interpretation is slow and prone to inter-observer variability. Deep learning models trained on large-scale ECG datasets face unstable convergence and degraded generalization due to non-stationary signal distributions and inter-patient variability. We propose GRU-CN, a Gated Recurrent Unit model with Continual Normalization, for four-class arrhythmia classification (AFIB, GSVT, SB, SR) on the 45,108-recording Chapman-Shaoxing dataset. GRU-CN uses a two-stage strategy: Stage 1 unsupervised GRU Autoencoder pre-training learns latent representations via encoder-decoder reconstruction; Stage 2 fine-tunes the pretrained encoder with a supervised classification head. Continual Normalization (CN) sequentially applies Group Normalization (GN) then Batch Normalization (BN) in the latent space, stabilizing training across intra- and inter-sample dimensions. Across five seeds (42–46), GRU-CN reaches a mean accuracy of 90.18% ± 4.17, macro F1 of 0.8909 ± 0.0463, and macro AUC of 98.17% ± 1.37, outperforming GRU-BN (88.92%), GRU-GN (88.06%), and a non-normalized baseline which was highly unstable across seeds (mean 67.47% ± 21.96, range 36.98–87.34%; McNemar’s test, p < 0.05 for all comparisons on a representative run). Ablation confirms autoencoder pretraining yields a substantial accuracy gain. Training loss is largely reproducible across seeds (std = 0.102), though one run converged to a lower local optimum, showing CN reduces but does not fully eliminate instability. Overall, combining CN with two-stage learning stabilizes training and improves arrhythmia classification on large-scale ECG data.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Athiya Jannati Zumaro
AU  - Najwa Muthia Aziz
AU  - Muhammad Nadhil Luthfirrahman
AU  - Dany Primanita Kartikasari
AU  - Rizal Setya Perdana
AU  - Kasyful Amron
AU  - Eko Sakti Pramukantoro
PY  - 2026
DA  - 2026/11/04
TI  - Arrhythmia Classification Using Continual Normalization-Based Gated Recurrent Unit on the Chapman-Shaoxing Dataset
BT  - Proceedings of the 2026 11th International Conference on Sustainable Information Engineering and Technology (SIET 2026)
PB  - Atlantis Press
SP  - 295
EP  - 304
SN  - 2589-4900
UR  - https://doi.org/10.2991/978-94-6239-828-3_26
DO  - 10.2991/978-94-6239-828-3_26
ID  - Zumaro2026
ER  -