Evaluating Different Graph Learning Techniques for Mental Task EEG Signal Classification

Abstract

Graph learning from the brain signals deals with capturing the changes in functional relationship between the brain regions during mental active and relaxed states. This paper investigates different graph learning techniques, namely geometry, signal similarity, and Graphical LASSO based methods for the classification of mental task from electroencephalogram (EEG) signals. Graph spectral energy based metric using Graph Signal Processing (GSP) technique is presented to classify mental active state from relaxed state. A binary KNN classifier is used to analyse each graph learning technique on publicly available Keirn and Aunon mental task EEG database. Performance of different graphs is then analysed and compared using classification Accuracy and F-Score.