Epileptic Seizure Detection using DWT based on MRMR Feature Selection Method

Epilepsy is classified as a dangerous neurological disorder with a high probability of death. People living with Epilepsy are advised to get appropriate treatment so that their condition does not worsen. Proper treatment is carried out using detection based on electroencephalogram (EEG) technology. EEG can help differentiate the condition of normal patients from patients with epilepsy. The EEG recording data used for this research was obtained online from the UCI Machine Learning Repository. There are five classes (seizure activity, non-seizure tumor brain area, non-seizure healthy brain area, non-seizure eyes closed, and non-seizure eyes open). The data obtained is designed and tested using a system to obtain good accuracy values. The EEG system design was created through several stages, including input dataset, decomposition, minimum, maximum, mean, standard deviation, kurtosis and skewness feature extraction from wavelet coefficients resulting from the DWT process selection of features MRMR and Support Vector Machine (SVM) classification. System testing is obtained from the K-Fold and Confusion Matrix processes. The best accuracy results show that MRMR succeeded in selecting three features with 95.30%, which were still able to provide performance almost equivalent to using 11 features with 96.03%, confirming that the three selected features were sufficient to represent epilepsy EEG signals well.