EUVIMEDEUVIMED entdecken
Crossref

A comparative study of deep neural network architectures with chaotic optimization for electroencephalography-based emotion recognition

Zahraa Tarek, Khaled Alnowaiser, Esraa Hassan

An International Journal of Optimization and Control: Theories & Applications (IJOCTA) · 21.05.2026 · 16 (3) · 869

DOI: 10.36922/ijocta026110037 ↗

Kurz gefasst

Fachartikel in der Zeitschrift An International Journal of Optimization and Control: Theories & Applications (IJOCTA), online erschienen am 21.05.2026. Verfasst von Zahraa Tarek und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Electroencephalography (EEG)-based emotion recognition has become an important research direction in affective computing, driven by its applications in mental health assessment, human--computer interaction, and brain--machine interfaces. Despite recent progress, challenges persist due to the high dimensionality of EEG features and the absence of consensus on optimal deep learning architectures for reliable emotion classification. This study introduces a unified comparative framework for EEG emotion recognition that combines statistical feature extraction with chaos-enhanced deep learning optimization. Five representative neural architectures, Transformer-based, CNN--LSTM hybrid, residual multilayer perceptron (MLP), capsule network, and hyperbolic-inspired network, were systematically evaluated under identical experimental conditions. EEG signals were normalized and transformed using advanced statistical feature extraction techniques, followed by robust data augmentation to improve model generalization. A chaos-driven learning rate optimization strategy based on the logistic map was incorporated into the training process to promote stable convergence and reduce susceptibility to local minima. Experimental results demonstrated that chaotic optimization substantially improved training stability and overall classification performance across architectures. The hyperbolic-inspired network achieved the highest validation accuracy of 93.21\% with a validation loss of 0.197 and a training time of 19.09 s, followed closely by the residual MLP, which attained 91.80\% accuracy with a loss of 0.226. The capsule network also showed competitive performance (90.87\% accuracy), whereas the CNN--LSTM hybrid exhibited comparatively lower accuracy (70.73%). These findings underscore the effectiveness of chaos-driven optimization and provide practical insights into architecture selection for efficient and robust EEG-based emotion recognition. The proposed framework offers a reproducible benchmark to support informed model design in affective computing applications.

Zitieren

APA

Tarek, Z., Alnowaiser, K., & Hassan, E. (2026). A comparative study of deep neural network architectures with chaotic optimization for electroencephalography-based emotion recognition. An International Journal of Optimization and Control: Theories & Applications (IJOCTA), 16(3), 869. https://doi.org/10.36922/ijocta026110037

Vancouver

Tarek Z, Alnowaiser K, Hassan E. A comparative study of deep neural network architectures with chaotic optimization for electroencephalography-based emotion recognition. An International Journal of Optimization and Control: Theories & Applications (IJOCTA). 2026;16(3):869. doi:10.36922/ijocta026110037

BibTeX
@article{Tarek2026,
  author = {Tarek, Zahraa and Alnowaiser, Khaled and Hassan, Esraa},
  title = {A comparative study of deep neural network architectures with chaotic optimization for electroencephalography-based emotion recognition},
  journal = {An International Journal of Optimization and Control: Theories & Applications (IJOCTA)},
  year = {2026},
  volume = {16},
  number = {3},
  pages = {869},
  doi = {10.36922/ijocta026110037}
}
Kostenlos starten →