Adaptive Flexible Analytic Wavelet Transform for EEG-Based Emotion Recognition

Video game development heavily relies on gaming emotions of players. Video games can trigger emotions that lead to hatred, aggressiveness, sadness, addiction, suicidal thoughts, etc. The effect of emotions can be reduced by studying the player's emotional state. The emotional state influences the psychological state. The electroencephalogram (EEG) signals generated due to neurological changes in the brain give accurate information about psychological states. This work introduces the adaptive flexible analytic wavelet transform (AFAWT) for detecting emotions using EEG signals. In AFAWT, parametric optimization finds the best basis function for representing EEG signals. Particle swarm optimization (PSO) is used to solve an inequality constraint problem, enabling the selection of appropriate AFAWT parameters. The AFAWT decomposes the EEG signal into subbands (SBs). The SBs' time-domain measures serve as features for classifying emotions in EEG signals. A post hoc multiple comparison analysis using the analysis of variance (ANOVA) test ensures the significance of the extracted features. Different classification algorithms test the obtained features for each SB. The hyperparameters of the classifiers neural network (NN), support vector machine (SVM), ensemble (EN), and k-nearest neighbors (k-NN) are optimized using ten-fold cross-validation and Bayesian optimization. Among the optimized classifiers, optimizable k-NN shows the best classification accuracy of 90.3% for four classes of emotions. Compared with other existing methods, our proposed method performs better on the same dataset.