An Approach of Human Emotional States Classification and Modeling from EEG

In this paper, a new approach is proposed to model the emotional states from EEG signals with mathematical expressions based on wavelet analysis and trust region algorithm. EEG signals are collected in different emotional states and some salient features are extracted through temporal and spectral analysis to indicate the dispersion which will unify different states. The maximum classification accuracy of emotion is obtained for DWT analysis rather than FFT and statistical analysis. So DWT analysis is considered as the best suited for mathematical modeling of human emotions. The emotional states are modeled with different mathematical expressions using the obtained coefficients from trust region algorithm that can be compared with the sub-band wavelet coefficients of different states. The proposed approach is verified with the adjusted R-square percentage and the sum of square errors. The adjusted R- square percentage of the mathematical modeled states are 78.4% for relax, 77.18% for motor action; however for memory, pleasant, enjoying music and fear they are 93%, 95.6%, 97.7% and 91.5% respectively. The proposed system is reliable that can be applied for practical real time implementation of human emotion based systems