Observed Shape Detection from EEG Time Series

Abstract

Brain computer interface studies required recording of a physiological response of a subject to exhibit relevant information. This extracted information can be used to perform an action and the amount of the information plays a significant role in the determination of brain computer interface (BCI) performance. The use of improved experimental paradigms as well as measuring the brain responses using electroencephalogram (EEG) is the most common approach for the BCI studies. In this study, the classification of the ongoing brain activity occurring as response to the four shapes is managed and reported. We applied Fourier transform to obtain the frequency spectrum regarding the one second time series of each channel with a time overlap of 50% to the feature set of each stimulus type. Four machine learning classifiers are implemented, and in the concept of the classification, (delta, theta, alpha, beta, and gamma) band power values for one second period constituted the feature set, resulting in a total of 315 features. Among the four ML classifier Quadratic Discriminant 87.1% recorded the highest accuracy.