A Genetic Algorithm Assisted Fuzzy Q-Learning epileptic seizure classifier

Abstract

Abstract Electroencephalograph is a technique of choice for detecting and analyzing various classes of epileptic seizures. In this work, reinforcement learning has been used to proactively classify epileptic seizures in an online manner. In particular, novel online Genetic Algorithm assisted Fuzzy Q-Learning and Fuzzy Q-Learning classifiers have been proposed for epileptic seizures. Proposed Reinforcement Learning based classifier uses Hilbert-Huang Transform to extract 19 time-frequency domain features in the pre-processing stage. Classification accuracy achieved with the proposed Genetic Algorithm assisted Fuzzy Q-Learning and Fuzzy Q-Learning approaches are 96.79% and 93.81%, respectively. This is comparable to the accuracy achieved by other contemporary seizure classification approaches and more accurate than other reinforcement learning based approach. The approach could serve as an effective tool in the hands of medical practitioners for analyzing bulk data and for speeding up seizure diagnosis.