Emotional stress detection using nonlinear and higher order spectra features in EEG signal

In this paper, an emotional stress recognition system from EEG signal is proposed. EEG is the reflection of brain activity and is widely used in clinical diagnosis and biomedical researches. The visual induction based acquisition protocol has been designed for recording the EEG signal, in six channels (FP1, FP2, T3, T4, O1 and O2), under two emotional stress states, Calm-neutral and Negative-exited for participants. The visual stimuli images are selected from the subset IAPS. In this research, we compare the dynamical properties of EEG signals with higher-order spectra analysis techniques. Chaotic invariants like correlation dimension, lyapunov exponent and fractal dimension are used to characterize the signal. We used genetic algorithm (GA) for optimum features selection for emotional stress classifier. The results show the average accuracy of 78.3% and 72.1% for two categories, with SVM and ENN classifiers, respectively. A comparison among the result shows that SVM classifer obtained better performance in emotional stress recognition.