Optimizing Weighted Permutation in Support Vector Machine for the Detection of Epilepsy via EEG Data Analysis

Epileptic seizures can range from short periods to violent shaking over long periods. Epilepsy is a disease that tends to occur repeatedly and cannot be cured. However, the occurrence of epileptic seizures can be controlled through medication. This study tried to detect epilepsy features via electroencephalogram (EEG) data analysis. The EEG data is divided into several windows using segmentation or decomposition. EEG Data will be extracted from each window using a modified Weighted Permutation Entropy (WPE) to produce one feature per window where the number of features of each EEG data will equal the number of windows obtained in the EEG data recording process. This research used k-fold cross-validation by dividing the data into training and testing data, classified using a Support Vector Machine. The EEG database used in this study came from Temple University Hospital EEG (TUH EEG), obtained online with as much as 1500 data. This EEG database consists of epileptic seizures (set 1) and non-epileptic seizures (sets 2, 3, 4, 5), each with 300 data. Based on testing, using this method produces an average accuracy of 81.29%.