Machine learning for seizure prediction: A revamped approach

Abstract

Occurrence of multiple seizures is a common phenomenon observed in patients with epilepsy: a neurological malfunction that affects approximately 50 million people in the world. Seizure prediction is widely acknowledged as an important problem in the neurological domain, as it holds promise to improve the quality of life for patients with epilepsy. A noticeable number of clinical studies showed evidence of symptoms (patterns) before seizures and thus, there is large research on predicting seizures. There is very little existing literature that systematically illustrates the steps in machine learning for seizure prediction, limited training data and class imbalance are a few challenges. In this paper, we propose a novel way to overcome these challenges. We present the improved results for various classification algorithms. An average of 21.71% improvement in accuracy is attained using our approach.