Feature Fusion Approach for Emotion Classification in EEG Signals

In this study, a novel approach is introduced for investigating the impact of diverse feature interactions on the categorization of emotional states within the context of human-computer interaction (HCI) using lateral electroencephalography (EEG). The approach presented in the study offers a unique method for emotion classification through the integration of numerous components. Feature extraction from preprocessed EEG recordings is executed judiciously using Differential Entropy (DE), Mean Square Teager-Kaiser Energy Operator (MST), and Sample Entropy (SampEn). Subsequently, the determination of the emotional state is conducted employing a Support Vector Machine (SVM) classification model. Experiments were carried out using the SEED-IV dataset, and the findings indicate that DE demonstrates the highest performance as a singular feature, with an average classification accuracy of 77.86%. Furthermore, a substantial improvement is achieved through the integration of non-linear Sample Entropy (SampEn) features with functional Minimum Spanning Tree (MST) attributes, resulting in an average classification accuracy of 84.58%. The incorporation of various characteristics and sophisticated electroencephalogram (EEG) analysis methods in the field of human-computer interaction (HCI) present a possible avenue for improving interactions between humans and computers.