Recognition of Anticipatory Behavior from Human EEG

Anticipation increases the efficiency of a daily task by partial advance activation of neural substrates involved in it. Single trial recognition of this activation can be exploited for a novel anticipation based Brain Computer Interface (BCI). In the current work we compare different methods for the recognition of Electroencephalogram (EEG) correlates of this activation on single trials as a first step towards building such a BCI. To do so, we recorded EEG from 9 subjects performing a classical Contingent Negative Variation (CNV) paradigm (usually reported for studying anticipatory behavior in neurophysiological experiments) with GO and NOGO conditions. We first compare classification accuracies with features such as Least Square fitting Line (LSFL) parameters and Least Square Fitting Polynomial (LSFP) coefficients using a Quadratic Discriminant Analysis (QDA) classifier. We then test the best features with complex classifiers such as Gaussian Mixture Models (GMMs) and Support Vector Machines (SVMs).