Deep Learning Based Emotion Detection Using Discrete Wavelet Transform

Emotion detection plays essential role in brain-computer interface, which applies in various fields such as healthcare, security, education. In this paper, a novel algorithm is proposed for detection of emotion using electroencephalogram signal. The method selects convolutional neural networks as the proposed classifier among different intelligent algorithms. Discrete wavelet transform is employed for signal decomposition into 4 bands namely theta, alpha, beta, and gamma, which are then used for feature extraction. A total of 800 features are chosen as the most relevant subset among 1920 features by recursive feature elimination algorithm in combination with 5-fold cross validation procedure and K-nearest neighbors model. The proposed algorithm is then validated with the selected feature subset on the validation set using 5-fold cross validation based statistic method. The performance results are accuracy of 65.5 % and 65.51 % for valence and arousal detection. Moreover, relative high F1_score values of 78.87% and 79.11 % for both valence and arousal classification, respectively, show potentiality of the proposed algorithm for recognition of emotion in practical environments.