Hybrid Supervised-Unsupervised Learning Pipeline for EEG Anomaly Detection Using Autoencoders and 1D CNN Models
Onkar Belure1, Ayush Awasthi2, Ankur Bhutare3
1Onkar Belure, Department of Computer Science and Engineering, MIT-ADT University, Pune (Maharashtra), India.
2Ayush Awasthi, Department of Computer Science and Engineering, MIT-ADT University, Pune (Maharashtra), India.
3Ankur Bhutare, Department of Computer Science and Engineering, MIT-ADT University, Pune (Maharashtra), India.
Manuscript received on 07 November 2025 | First Revised Manuscript received on 26 April 2026 | Second Revised Manuscript received on 04 May 2026 | Manuscript Accepted on 15 May 2026 | Manuscript published on 30 May 2026 | PP: 23-25 | Volume-14 Issue-6, May 2026 | Retrieval Number: 100.1/ijese.B471515021225 | DOI: 10.35940/ijese.B4715.14060526
Open Access | Editorial and Publishing Policies | Cite | Zenodo | OJS | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open-access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Seizure detection from electroencephalogram (EEG) signals remains a difficult problem because of the wide variation in patient patterns and the limited amount of labelled data. In this work, we developed a hybrid learning setup that blends a supervised 1D Convolutional Neural Network (CNN) with an unsupervised Autoencoder (AE). The CNN learns to recognise seizure-related patterns from labelled EEG segments, while the AE models typical EEG activity and signals abnormal deviations. We combined their predictions using three ensemble techniques-Soft Weighted Average, Soft Average, and Majority Voting-to stabilize performance and reduce false alarms. Tests on the Turkish Epilepsy EEG Dataset showed that this hybrid approach performed more reliably across patients than either model alone.
Keywords: EEG, Seizure Detection, Autoencoder, 1D CNN, Hybrid Learning, Ensemble Learning, Anomaly Detection, Biomedical Signal Processing.
Scope of the Article: Computer Science and Engineering