Automatic Prediction of Epileptic Seizure Using Kernel Fisher Discriminant Classifiers

Accurate classification of seizure and non-seizure EEG signals is an important step in epileptic seizure prediction. In this paper, a seizure prediction algorithm based on kernel Fisher discriminant (KFD) classifiers is proposed. In this algorithm, spectral features are extracted by wavelet transform from seizure and non-seizure EEG signals. Then an efficient leave-one-out cross-validation of KFD classifier is used to classify the extracted features from EEG signals. This classifier have a low computational complexity that being significantly faster than conventional k-fold cross-validation procedures and being an attractive means of model selection in large-scale applications. The performance of algorithm is evaluated based on four measures, accuracy, false detection rate (FDR), good detection rate (GDR) and delay. The results illustrate that the algorithm can recognize 81 seizures of all 87 seizures with average delay of 3.7 second.