Ensemble Deep Learning Technique for Robust Seizure Detection Using Integrating Convolutional and Recurrent Neural Networks with Advanced Optimization Techniques
International Journal of Basic and Applied Sciences · 2025
DOI: 10.14419/04tp2d60 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Basic and Applied Sciences, 2025. Verfasst von K. Dileep Kumar und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Electroencephalogram (EEG) is a common problem where seizures need to be detected in the healthcare industry because the quality of life of a patient can be affected, and it can even cause serious health issues. In this paper, a hybrid deep learning model combining Convolutional Neural Networks (CNN) to capture spatial features and Long Short-Term Memory (LSTM) RNNs to detect temporal dynamics in EEG signals is presented. Our model was trained and tested on the CHB-MIT Scalp EEG Database, which has been used in many studies on seizure detection. The hybrid CNN-LSTM structure does better than the conventional CNN or LSTM structure and its accuracy is 94.3 %, its precision score is 92.9 and it has a high recall score of 99.2 with F1 score of around CNN-only model, however, performed a little better gaining 89.5 percent of accuracy along with only approximately 86.6 percent of increase in F1 score. The LSTM has also performed badly with a range of 84-86%. Moreover, after conducting the study, the research also takes a step further in determining how robust these models are against adversarial attack by attacking them with attacks using Fast Gradient Sign Method (FGSM). At 0.05, the accuracy was reduced which is expected to happen to all models but the hybrid model had a lesser impact hence returning with 78:5% inaccuracy. The hybrid model (0.833) had better robustness score compared to the CNN-only and the LSTM-only model as in the case of main model results with MaxFillup. Adversarial training also plays a critical role in enhancing the underpinnings of this hybrid model such that when it was tested on clean data, it was tested robustly against white-box attacks. The hybrid model proposes a possible approach toward real-time seizure detection in clinics, which besides being more accurate can be adversarailly robust as well. The contributes to the current body of literature on seizure detection a fully integrated approach toward robust behavior in practical medical environments.
Zitieren
APA
Kumar, K. D., Dash, D. S., & A., D. R. K. G. (2025). Ensemble Deep Learning Technique for Robust Seizure Detection Using Integrating Convolutional and Recurrent Neural Networks with Advanced Optimization Techniques. International Journal of Basic and Applied Sciences. https://doi.org/10.14419/04tp2d60
Vancouver
Kumar KD, Dash DS, A. DRKG. Ensemble Deep Learning Technique for Robust Seizure Detection Using Integrating Convolutional and Recurrent Neural Networks with Advanced Optimization Techniques. International Journal of Basic and Applied Sciences. 2025. doi:10.14419/04tp2d60
BibTeX
@article{Kumar2025,
author = {Kumar, K. Dileep and Dash, Dr. Sachikanta and A., Dr. Rajendra Kumar Ganiya},
title = {Ensemble Deep Learning Technique for Robust Seizure Detection Using Integrating Convolutional and Recurrent Neural Networks with Advanced Optimization Techniques},
journal = {International Journal of Basic and Applied Sciences},
year = {2025},
doi = {10.14419/04tp2d60}
}
EUVIMED