Seizure forecasting

The focus of this project is to predict impending seizure occurrence for epilepsy patients using intracranial EEG recordings. Reliable prediction of seizures can provide the patients with not only warnings but also new therapeutic possibilities. The goal is to correctly differentiate the pre-seizure brain state (the preictal state) from the baseline state (the interictal state) using just the iEEG signals. Both frequency domain and time domain features have been extracted and explored and ML algorithms such as logistic regression, lasso logistic regression, svm and knn have been applied. A kaggle score of 0.77486 was achieved using the logistic regression and frequency domain features.