Comprehensive Summary
This study aims to investigate a new method for predicting epileptic seizures with accuracy through EEG signal detection. This study utilized the CHB-MIT dataset. The experimental method began with EEG signal preprocessing to eliminate noise with a Butterworth filter and a wavelet, followed by feature extraction using customized one-dimensional CNN, and finally signal classification through a hybrid classifier system involving random forest (RF), long short-term memory (LSTM), and support vector machines (SVM). The method, after training, resulted in 98.67% specificity, 99.34% sensitivity, and a false positive alarm rate of 0.039. This epileptic seizure prediction method achieved more accurate outcomes than previously established methods.
Outcomes and Implications
With outcomes more accurate than that of traditional methods, the proposed method is an improved alternative to predicting epileptic seizures. Further work will be conducted in an attempt to build in other vital signs for a more all-encompassing analysis when predicting epileptic seizures. Additionally, the model will be modified to consist of a more lightweight, practical interface for future applications.