Seizure Prediction With Spectral Power of EEG Using Cost-Sensitive Support Vector Machines