Prediction of decision-making response in ultimatum game by constructing regularized common spatial network pattern based on phase locking value

Abstract

Decision-making is a complex cognitive process and plays an important role in the interaction between people. Many researchers are striving to predict the individual's decision-making response(ie., acceptance or rejection) by processing electroencephalogram(EEG) trial-by-trial. In the study, we proposed a supervised learning approach, called regularized discriminative spatial network pattern(RDSNP), to predict individual responses with a small size of training data set. It constructs discriminative brain networks by calculating the phase lock value of different decision-making responses with single-trial EEG data. Then the single-trial spatial network topology was applied to extract the RDSNP features. Finally, a linear discriminate analysis(LDA) classifier was built on RDSNP features and used to predict individual decisions trial-by-trial. To verify the performance of RDSNP, we compared this approach with such widely used baseline feature extraction methods as event related potentials, network properties, principal component analysis in EEG signals of 16 subjects, which was acquired during the experiments of ultimatum game, in terms of accuracy and F1-score suggests that our approach achieve a better performance on predicting single-trial decisions.