Spectrum estimation of odor EEG responses with parametric-nonparametric spectral analysis methods

It is known that external stimulus such as visual, auditory and odor have effect on brain activity. Effects of odor stimuli, which has complex structure, on central nerveous system is lack of knowledge in literature. The goal of proposed study is to show how pleasant-unpleasant odors effect brain waves by using spectral analysis methods and discriminating different odors through statistical methods and a classifier. The EEG dataset used in study was taken from 6 participants while their eyes are closed and 4 odor (2 pleasant-2unpleasant) stimulus were applied to them using 14 chanelled EMOTIV-EPOC headset. Discrete Wavelet Transform (DWT) was used to pre-processed signals obtained from embedded filters to extract more meaningful EEG sub-bands (delta-tetha-alpha-beta). First of all, power spectrum graphics of these sub-bands was drawn using Welch's method to see how pleasant-unpleasant odor EEGs behave. Then, spectrum coefficients were gained by help of parametric (Burg, Yule-Walker, Covariance, Modified Covariance) and non-parametric (Welch's) methods. Selected feature vectors from these coefficients were classified. Selected features are min, max value and standard deviation. k-NN was chosen for classification algorithm. Avarage power spectrum analysis showed that unpleasant odor EEG has higher values than pleasant one with respect to all sub-bands. Parametric methods gave better results to discriminate odor EEGs. Burg method has highest classification rate among others.