Machine Learning-Based Automatic Emotion Classification Using EEG Signals

Today, a wide variety of studies are carried out on the automatic interpretation of brain signals. At the beginning of these studies are emotion recognition and classification. Emotions play an active role in individuals' relations with the outside world, their actions and decisions. Therefore, emotion recognition is critical in human-computer interaction. It is a fact that various activities develop in the brain in the emotional changes of individuals. At this point, electroencephalogram (EEG) signals provide important clues. In this study, an approach using local binary patterns and statistical moment methods is proposed for automatic emotion recognition and classification. In this model, tunable Q wavelet transform method is used for signal separation and feature extraction is performed from both the raw signal and the subbands of this signal. ReliefF, a well-known method in the literature, was used in the feature selection phase of the model, and the most significant features were selected with this method. In the classification phase of the model, support vector machines, a classical classifier, were used. In this study, which exemplifies a basic level machine learning model, the DREAMER dataset is tested to validate the proposed model, and the k-fold cross-validation strategy is applied as a validation technique. In this study, in which arousal, valence and dominance states were examined, binary classification was performed