Investigation of Human Emotion Pattern Based on EEG Signal Using Wavelet Families and Correlation Feature Selection

Abstrak

Emotions is one of the advantages given by God to human beings compared to other living creatures. Emotions have an important role in human life. Many studies have been conducted to recognize human emotions using physiological measurements, one of which is Electroencephalograph (EEG). However, the previous researches have not discussed the types of wavelet families that have the best performance and canals that are optimal in the introduction of human emotions. In this paper, the power features of several types of wavelet families, namely Daubechies, symlets, and coiflets with the Correlation Feature Selection (CFS) method to select the best features of alpha, beta, gamma and theta frequencies. According to the results, coiflet is a method of the wavelet family that has the best accuracy value in emotional recognition. The use of the CFS feature selection can improve the accuracy of the results from 81% to 93%, and the five most dominant channels in the power features of alpha and gamma band on T8, T7, C5, CP5, and TP7. Hence, it can be concluded that the temporal of the left brain is more dominant in recognition of human emotions.