Computationally Efficient Epileptic Seizure Prediction based on Extremely Randomised Trees

Sheng Wong, Levin Kuhlmann

Abstract: Epilepsy is a neurological disorder that affects close to 60 million of the world's population and is commonly categorized by having unpredictable seizure episodes. Over the years, in attempt to predict epileptic seizures in patients using electroencephalographic (EEG) data, several machine learning based models and algorithms have been developed but many of them present shortcomings such as having computationally inefficient algorithms, limited EEG data and there is no one size fits all patients model. Here a generalised seizure prediction algorithm based on extremely randomised tree classification is presented that can be applied to all patients with a minimal number of features to provide increased computational efficiency and comparable performance score relative to a more complicated state-of-the-art algorithm. The new algorithm achieves a 3.25 factor speed up in computation time while achieving an average Area under the curve, AUC of 0.74 relative to 0.72 for the state-of-the-art algorithm. The algorithm is designed to be implemented on small implantable/wearable EEG devices with little computing power, in order to preserve battery life and help make seizure prediction a clinically viable option for patients with epilepsy.