.png){kind=icon}
.png){kind=icon}
Return
Epilepsy Seizure Prediction sing an SVM Algorithm
.png){kind=icon}
PRE
AI
DOI：10.54364/AAIML.2026.61281.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Epileptic seizures remain a real concern and a major medical challenge, simply because they strike unexpectedly and without warning, turning a patient's life upside down. Undoubtedly, the ability to detect an impending seizure well in advance is a lifeline, aiding physicians and completely transforming the course of treatment. However, there is a hurdle: while complex artificial intelligence models (such as deep learning) are highly accurate, they are resourceintensive and require powerful computers, making them difficult to run on small or portable devices. Therefore, in this study, we developed a smarter and lighter solution: a model based on the SVM algorithm. The idea behind this model is that it focuses specifically on the two minutes preceding a seizure, making it lightweight and easy to implement even on devices with limited capabilities. We analyzed brain signals and extracted the necessary data, and using this model, we were able to clearly distinguish between normal brain activity and the moments before a seizure. The results were very promising. We achieved an accuracy rate of nearly 80%, with the ability to provide warnings of an impending seizure 5 to 10 minutes before it occurs. We observed that delta and gamma waves were the most effective in detecting this threat. In short, what distinguishes this approach from others is its simplicity and low resource consumption compared to other complex systems, making it ideal for use on wearable devices (such as medical watches) and in real time. We plan to test this model on a larger scale, including data from adult patients at different centers, to ensure its effectiveness for all patients.
Keywords:
Epilepsy
EEG
Seizure Prediction
Support Vector Machine
Frequency Do-main
Wavelet Transform
Preictal Detection
Machine Learning
Journal
A
IF:
0.5
Papers:
44
Citations:
0