On the temporal behavior of EEG recorded during real finger movement

Abstract

This paper studies the temporal behavior of EEG data during self-paced real finger movement using Hidden Markov Models and Conditional Random Fields and proposes novel temporal classification methods for movement classification versus idle state. Results are compared to those from Linear Discriminant Analysis based classification. It is demonstrated that using the temporal information in the classification model itself can significantly improve the performance of self-paced EEG classification. The proposed methods are tested on 15 subjects, achieving between 57% and 88% cross validation accuracy, with an average 6% improvement in classification accuracy. © 2011 Springer-Verlag.