Seizure prediction algorithm based on simulated annealing and machine learning

Epilepsy is one of the widespread diseases of the central nervous system around the world, which is characterized by seizures with different periods and symptoms among people. Expecting incoming seizures is important and necessary to organize the patient's life and take the necessary precautions to preserve his life. We used a seizure prediction algorithm based on four stages. In the first round, one of the person's Electroencephalogram (EEG) channels is selected to be the input channel based on a statistical analysis of the person's EEG recordings. In the second round, the preictal period length (PIL) and the length of the sample segment (SEG) used are calculated. In the third round, several features from the selected channel’s data are extracted, while in the fourth round, the simulated annealing method is used to find a set of features that achieve the best performance in differentiating between the Preictal and Interictal periods. The results showed an average sensitivity of 77\% and a low false prediction rate of 0.2 by testing the algorithm with the CHB-MIT scalp dataset. Conclusion and Significance: Comparing our results with the results of recent research showed the superiority of our results in terms of using one channel compared to multiple channels in other research.