Epileptic Seizure Detection Using Machine Learning and Deep Learning Method

Abstract

Seizures are a common symptom of epilepsy, a nervous system disease. Epilepsy can be detected with an Electroencephalogram (EEG) signal that records brain nerve activity. Visual observations cannot be done on a routine basis because the EEG signal has a large volume and high dimensions, so a method for dimension reduction is needed to maintain signal information. Appropriate features should be selected to reduce computational complexity and classification time in detecting epileptic seizures. This study compares the performance of Machine Learning and Deep Learning models to detect epileptic seizures to get the best performing model. The feature extraction process using Discrete Wavelet Transform (DWT) taking feature values, namely maximum, minimum, standard deviation, mean, median, and energy. Furthermore, feature selection uses correlation variables, namely removing uncorrelated variables using threshold variations. The improvement of this study is to use six features, namely the maximum, minimum, standard deviation, mean, median, and energy values, as input values in the classification process. Non-seizure signals and epileptic seizures were classified using Machine Learning: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), Decision Tree (DT), and Deep Learning: Long Short-Term Memory (LSTM). The trials used three variations of datasets, namely dataset 1: 96 signals, dataset 134 signals, and dataset 3: 182 signals. Nine different classification experiments were conducted using four performance evaluation indicators: accuracy, precision, recall, and F1-Score. Based on the test results, the model with the best performance is the SVM method with 100% accuracy, 100% precision, 100% recall, and 100% f1-score.