Prediction of the Brain Surprise in Sequences of Oddball Stimuli

Certain features of the Event Related Potential (ERP) signal such as the P300 and N200 components are known to reflect the surprise of the brain when encountering unlikely samples in a sequence of stimuli. In this paper, we propose a method to learn the amplitude of these components for an arbitrary random stimuli sequence. First, a Fisher optimization method is developed for binary prediction of the P300 occurrence. Then, a Generalized Linear Method (GLM) is employed to predict the amplitude of the P300 component. The proposed method is applied to an auditory oddball experiment with recorded Electroencephalogram (EEG) data. Both developed methods are shown to yield promising results with an acceptable prediction error.