An Energy Efficient Seizure Prediction Algorithm

In this project, we sought to develop a learning algorithm that identifies EEG time series as pre-ictal (the time just before a seizure occurs) or inter-ictal (the time between seizures) using undersampled data to reduce the energy consumption and bandwidth usage. By training the model with segmented 1 second intervals and predicting seizures with a decision window, we achieved high seizure prediction accuracy at a downsampling rate as high as 40. These techniques can be broadly applied to reduce energy demands for a variety of wearable medical and health device applications.