A Comparative Study of Deep Learning Algorithms for Epileptic Seizure Classification

Epilepsy is a potentially lethal and widespread common neurodegenerative illness that affects millions of people worldwide. Electroencephalography (EEG) is a test used to evaluate the brain’s electrical activity, and one of its applications is used to detect epilepsy seizures. A trained neurologist does a manual inspection of EEG, but it is an extensive and burdensome process, thus affecting the performance of the neurologist. This proposed work has reached various deep RNN based models like LSTM, GRU, Bidirectional LSTM, and Bidirectional GRU based on accuracy, sensitivity and specificity, and f-1 score. We performed our model on the Bonn data set. Moreover, we have found out that Bi-directional LSTM works best for detecting epilepsy seizures in binary class with 99.33% accuracy, ternary epilepsy case with 98%accuracy, and lastly, for five class detection with 82% accuracy. Thus, these RNN based models gave good results on smaller EEG signals so that they could alert patients about their incoming epileptic seizures.