Towards Superior EEG-Based Emotion Recognition: Integrating CNN Outputs with Machine Learning Classifiers for Enhanced Performance

Emotions exert a foundational influence on human experiences, influencing daily interactions, decision-making processes, and overall well-being. The integration of electroen-cephalography (EEG) into emotion recognition has emerged as a crucial component in advancing affective computing and enhancing Human-Computer Interaction (HCI). This study makes a noteworthy contribution to existing methods for EEG-based emotion identification, leveraging the SEED dataset and introducing distinctive features, particularly Differential Entropy (DE). Initiating with the training of a 2D Convolutional Neural Network (CNN) using the DE feature, a commendable accuracy rate of 89.0% on the SEED dataset was achieved. Subsequently, the prediction probabilities generated by the 2D CNN were employed to construct a new feature vector for each sample. Utilizing this vector, a novel feature map was created, and a range of machine learning (ML) classifiers including decision tree (DT), random forest (RF), K-nearest neighbors (KNN), and support vector machine (SVM) were independently trained. These diverse classifiers were then amalgamated through a soft voting-based approach to augment the overall classification performance, resulting in an impressive 93.33 % accuracy. The experimental outcomes underscore the effectiveness of the proposed methodology, showcasing superior performance when compared to various approaches in the realm of EEG-based emotion analysis. This research contributes substantially to the ongoing evolution of emotion recognition technology, fostering a deeper understanding of human-computer interaction.