Epilepsy is a neurological disorder characterized by recurrent seizures. Automated prediction of epileptic seizures is essential in monitoring the health of an epileptic individual to avoid cognitive problems, accidental injuries, and even fatality. In this study, scalp electroencephalogram (EEG) recordings of epileptic individuals were used to predict seizures using a configurable Extreme Gradient Boosting (XGBoost) machine learning algorithm. Initially, the EEG data was preprocessed using a standard pipeline. We investigated 36 minutes before the onset of the seizure to classify between the pre-ictal and inter-ictal states. Further, temporal and frequency domain features were extracted from the different intervals of the pre-ictal and inter-ictal periods. Then, the XGBoost classification model was utilized to optimize the best interval for the pre-ictal state to predict the seizure by applying Leave one patient out cross-validation. Our results suggest that the proposed model could predict seizures 10.17 minutes before the onset. The highest classification accuracy achieved was 83.33 %. Thus, the suggested framework can be optimized further to select the best features and prediction interval for more accurate seizure forecasting.
Keywords: Automated Detection; Electroencephalogram; Epilepsy; Machine Learning; Seizure Prediction.