Automated Classification of Epileptic Seizures from EEG Signals Using Deep Neural Network

Seizures are caused by the chronic brain condition epilepsy. Epilepsy syndromes come in a variety of forms, each with their own distinctive features. The electroencephalogram (EEG) gauges brain electrical activity. Machine learning is used by automated epilepsy seizure detection (AESD) systems to examine EEG signals and spot seizures. This paper uses machine learning techniques to provide a unique method for the automated identification of epileptic episodes. We employ cutting-edge deep learning algorithms and a large dataset of EEG signals to precisely identify seizure and non-seizure occurrences. A long short-term memory recurrent neural network model was implemented for detecting epileptic seizures using EEG time series data. The 178-dimensional input was normalized and sequenced prior to training. The architecture incorporated dropout regularization and dense layers before a 5-unit output layer. Across 100 epochs and 15 batch size, 89.7% test accuracy was attained. An 8-layer feedforward network was also explored, achieving 98.5% accuracy. The final trained LSTM model loaded from saved weights reached 99.9% accuracy on the test set. These results demonstrate deep learning's capability for automated seizure classification through mining complex patterns in EEG signals. Further optimization of network architecture and hyper-parameters could potentially boost accuracy. Overall, the study highlights promising avenues for developing noninvasive AI-based diagnostic tools for epilepsy using scalp EEG data. When it came to predicting epileptic seizures from EEG data, the long short-term memory (LSTM) recurrent neural network architecture presented in this work had the highest accuracy of all the models tested, coming in at 99.9%. Compared to the next best performing model, an artificial neural network (ANN), which achieved an accuracy score of 98.5%, this signifies a 1.4% improvement.