EEG-Based Seizure Prediction Using CNN and BiLSTM Algorithm of Deep Learning

Automated seizure identification is critical for evaluating epilepsy because manually reviewing electroencephalogram (EEG) data is time- consuming and error-prone. Recent advances in deep learning allowed the creation of automatic seizure identification systems that may evaluate EEG raw time series recordings without requiring the extraction of features manually. The data is divided into EEG epochs and labelled into five categories, including activity related to seizures. Here concentrate on two tasks evaluating seizure activity versus normal activity and categorizing seizure action against every other category. Here two DL algorithms are (CNN) convolutional neural network and a (Bi-LSTM) Bi-directional long short-term memory. These algorithms are taught to divide input EEG data into groups that have been identified. The CNN model is intended to extract significant characteristics from EEG time series data using layers of convolution, followed by pooling layers. However, CNNs might fail to recognize long-term dependencies found in continuous information such as electroencephalogram recordings. On the other hand, the BILSTM model may capture temporal dependencies in sequential data by exploiting both the forward and backward flow of information by using LSTM neurons. This design works well with data from time series, such as EEG recordings, where identifying temporal patterns is critical for accurate categorization. After training and testing the models on the dataset, the prediction results show that the BILSTM algorithm performs much better than the CNN model i both tasks. The BILSTM Model achieves an accuracy of above 98.15%, indicating its effectiveness in accurately classifying seizure activity from EEG recordings compared to the CNN model.