A novel similarity-constrained feature selection method for epilepsy detection via EEG signals
Abstract Epilepsy constitutes a persistent neurological disorder characterized by recurrent paroxysmal neuronal hyperactivity. The automatic recognition of epilepsy by electroencephalography (EEG) holds significant value for epilepsy treatment and medical diagnosis. Current methods for epilepsy EEG...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-07-01
|
| Series: | Journal of King Saud University: Computer and Information Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s44443-025-00152-w |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Epilepsy constitutes a persistent neurological disorder characterized by recurrent paroxysmal neuronal hyperactivity. The automatic recognition of epilepsy by electroencephalography (EEG) holds significant value for epilepsy treatment and medical diagnosis. Current methods for epilepsy EEG signal recognition typically extract and represent features from multiple dimensions, such as temporal, spectral, and time–frequency analysis. However, excessive feature extraction can increase model complexity and reduce generalization performance. To address this issue, we develop a novel EEG feature selection approach, which considers both intra-class and inter-class similarities to minimize the feature set size while preserving model effectiveness. First, the notion of sample similarity is introduced, and intra-class similarity and inter-class similarity are defined. Then, an optimization problem for feature selection is formulated by enhancing intra-class similarity and reducing inter-class similarity. Finally, a heuristic search strategy-based algorithm is designed to select features for epileptic EEG signals. This approach incorporates cross-validation and constructs locally optimal feature sets to minimize feature dimensionality, enhance model adaptability across different datasets, and enhance stability, thereby increasing the generalization capability of the detection framework. Experimental validation was conducted using the Bonn EEG dataset, with benchmark tests revealing the proposed framework's superior feature consistency and classification accuracy compared to conventional approaches. |
|---|---|
| ISSN: | 1319-1578 2213-1248 |