Analysis of the EEG Signal Using Higher-Order Spectra (HOS) in the Neuro-marketing Application

Understanding how consumers make decisions is one of the topics of interest for researchers, marketers, and industry managers. In the present study, the electroencephalogram (EEG) signals of 25 participants were used while viewing 14 different products. First, the EEG signals were pre-processed and then the high-order spectrum (HOS) features as the sum of the size of the Bispectrum, the sum of the square size of the Bispectrums, the sum of the size of the Bicoherences, and the sum of the square size of the Bicoherences in each of the 10 frequency ranges as well as the whole frequency range and features of Heinich test such as Chi-square value (CSV), probability of false alarm (Pfa), and Lambda were extracted to investigate the relationship between product liking and dislike. A total of 48 features were calculated for each EEG channel. By calculating them in 14 channels, 672 features were obtained for each sample. The superior traits were selected using a genetic algorithm (GA) and the nearest neighbor method in the wrapper model. The traits were classified using multi-layer perceptron neural network (MLP) and support vector machine (SVM). In the feature selection stage, 206 features were obtained. The results of the study showed that the proposed model with the help of SVM with Gaussian kernel can reach an average accuracy of 73.24% on all users. The proposed method, thus, seems to have a satisfactory performance in identifying the likes and dislikes of the product and could be useful in the neuro-marketing application.