← Back to Publications List

Attention-Enhanced BiLSTM Autoencoder for ECG Anomaly Detection

Students & Supervisors

Student Authors
Abu Nayem Md Arman
Bachelor of Science in Computer Science & Engineering, FST
Sheikh Syeed Ul Haque
Bachelor of Science in Computer Science & Engineering, FST
Shammam Raiyan
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Kamruddin Nur
Professor, Faculty, FST
Rifat Al Mamun Rudro
Lecturer, Faculty, FST

Abstract

The research work introduces an unsupervised deep learning architecture, specifically, the attention-based BiLSTM Autoencoder algorithm, for the purpose of detection of abnormal cardiac signals using ECG signals without the use of any labeled data for abnormal heartbeats. The system uses CNN layers for local feature extraction, BiLSTM layers for sequence learning, self-attention for identifying critical regions of the waveform, and temporal smoothing to minimize false positives. The experiment performed on the ECG5000 dataset revealed good performance by the method in terms of ROC-AUC of 0.9531 and recall of 0.9860.

Keywords

ECG anomaly detection unsupervised learning BiLSTM autoencoder self-attention biomedical signal analysis.

Publication Details

  • Type of Publication:
  • Conference Name: International Conference on Power, Electronics, Communications, Computing, and Intelligent Infrastructure 2026 (PECCII2026)
  • Date of Conference: 17/06/2026 - 17/06/2026
  • Venue: Zoom/Pabna University of Science and Technology, Jhenaidah, Bangladesh
  • Organizer: Faculty of Engineering and Technology at Pabna University of Science and Technology (PUST).
View on Publisher