A nonlinear dynamical approach to analysis of emotions using EEG signals based on the Poincaré map function and recurrence plots

Abstract

AbstractDynamic variations of electroencephalogram (EEG) contain significant information in the study of human emotional states. Transient time methods are well suited to evaluate short-term dynamic changes in brain activity. Human affective states, however, can be more appropriately analyzed using chaotic dynamical techniques, in which temporal variations are considered over longer durations. In this study, we have applied two different recurrence-based chaotic schemes, namely the Poincaré map function and recurrence plots (RPs), to analyze the long-term dynamics of EEG signals associated with state space (SS) trajectory of the time series. Both approaches determine the system dynamics based on the Poincaré recurrence theorem as well as the trajectory divergence producing two-dimensional (2D) characteristic plots. The performance of the methods is compared with regard to their ability to distinguish between levels of valence, arousal, dominance and liking using EEG data from the “dataset for emotion analysis using physiological” database. The differences between the levels of emotional feelings were investigated based on the analysis of variance (ANOVA) test and Spearman’s statistics. The results obtained from the RP features distinguish between the emotional ratings with a higher level of statistical significance as compared with those produced by the Poincaré map function. The scheme based on RPs was particularly advantageous in identifying the levels of dominance. Out of the 32 EEG electrodes examined, the RP-based approach distinguished the dominance levels in 23 electrodes, while the approach based on the Poincaré map function was only able to discriminate dominance levels in five electrodes. Furthermore, based on nonlinear analysis, significant correlations were observed over a wider area of the cortex for all affective states as compared with that reported based on the analysis of EEG power bands.