No abstract

More than 65 million people live with epilepsy. The unpredictable nature of epileptic seizures drastically increases the risk of injury, especially in daily activities such as walking or driving. The purpose of this project is to develop an accurate prediction device that utilizes raw EEG data for the prediction of epileptic seizures to alert patients of an oncoming seizure beforehand to escape dangerous situations. Using the raw EEG data, features were extracted by computing the average power spectral density of different brain waves after applying the Fast Fourier Transform. These features were used as the input dataset to the machine learning algorithms. Each model is tested with new unseen data using various metrics such as accuracy, precision, recall, and F1 score. The highest performing algorithm, Random Forest (RF) produced a prediction accuracy of 99.0% and a precision of 99.3%. Channel importance is calculated for the RF algorithm. This analysis helped to reduce the number of channels from 22 before feature importance to only 7 channels without significant hits to performance metrics. Using the RF algorithm, an embedded program is developed to run on a portable, low-power hardware device to predict the onset of a seizure. The hardware includes BeagleBone Black microcontroller running open-source software and a Bluetooth transmitter-receiver to transmit the prediction to smartphone devices. By reducing the number of EEG channels to 7 channels, the system is more convenient for a future wearable device. Hardware with the ability to predict epileptic seizures can save many patients from potentially dangerous situations such as driving or swimming. It can help many patients in their daily lives by removing the uncertainty and improving their quality of life.