Comparing Different Classifiers and Features for Electroencephalography-Based Product Preference Recognition

Abstract

Brain activity analysis during the visualization of different commercial images can help better understand brain activities and their application in neuromarketing. This study evaluates different electroencephalography (EEG) time- and frequency-domain features within different brain regions with six different classifiers, namely, k-nearest neighbors, pseudo-quadratic discriminant analysis, na¨ıve Bayes, support vector machine (SVM), random forest (RF), and decision tree, to determine the best features and brain regions associated with decision-making. An online dataset of 25 users’ responses to 42 products using a 14-channel EEG system was used. The outputs included two classes: like and dislike. Twenty-one features were derived from the preprocessed data using a window size of 1 s for 4 s for the EEG signals. The best-performing classifiers were SVM and RF, and the best features were Willison amplitude (66.9%) and Hjorth complexity (66.3%) using all channels. Furthermore, the temporal and frontal lobes of the brain showed higher accuracy than other regions, and the right frontal lobe was more dominant than the left frontal lobe in relation to product preference decisions and displayed the potential to classify users’ decisions for future simplified systems.