Epileptic seizure prediction using wavelet transform, fractal dimension, support vector machine, and eeg signals

Epilepsy, a neurological disorder, affects millions of persons worldwide. It is distinguished by causing recurrent seizures in patients, which can conduct to severe health problems. Consequently, it is essential to offer a method capable of timely predicting a seizure before its appearance, so patients can avoid possible injuries by taking preventive action. In this sense, a method based on the integration of discrete wavelet transform (DWT), fractal dimension, and support vector machine (SVM) is presented for the prediction of an epileptic seizure up to 30 min before its onset through the analysis of electroencephalogram (EEG) signals. DVVT is initially applied to the EEG signals to obtain their main neurological bands; then, five fractal dimension indices (e.g. Sevcik, Petrosian, Box, Higuchi, and Katz) arc explored as potential seizure indicators. Finally, an SVM is developed to predict the epileptic seizure automatically. The effectiveness of the proposal to predict an epileptic crisis is validated by employing a database of 14 subjects with 42 epileptic seizures provided by the Massachusetts Institute of Technology and the Children's Hospital Boston. The results demonstrate that the proposal can predict an epileptic seizure up to 30 min before its onset with a high accuracy of 93.33%.