Significant preprocessing method in EEG-Based emotions classification

EEG preprocessing methods for classifying person emotions have been widely applied. However, there still remain some parts where determining significant preprocessing method can be improved. In this regards, this paper proposes a method to determine the most significant preprocessing methods, among them to determine (i) denoising method; (ii) frequency bands; (iii) subjects; (iv) channels; and (v) features. The purposes are to improve the accuracy of emotion classification based on valence and arousal emotion model. EEG data from 34 participants will be recorded with the questionnaires (valence and arousal) that have been taken from the participants when they receive stimuli from picture, music, and video. EEG data will be divided into 5 seconds for each trial. Then, EEG data will be processed using denoising method and feature extraction. After that, the most significant preprocessing methods will be chosen using statistical analysis Pearson-Correlation. The preprocessed EEG data will be categorized. The average accuracy results using SVM are 66.09% (valence) and 75.66% (arousal) while the average accuracy results using KNN are 82.33% (valence) and 87.32% (arousal). For comparison, the average accuracy results without choosing the most significant preprocessing method are 52% (valence) and 49% (arousal) using SVM while the average accuracy results using KNN are 50.13% (valence) and 56% (arousal).