Enhanced Techniques for Multi-Emotion Detection from EEG Signals: Focus on Channel and Feature Selection

This paper presents a comprehensive framework for multi-emotion detection using electroencephalogram (EEG) signals, focusing on feature extraction, channel selection, feature selection, and classification. The proposed methodology leverages advanced techniques such as Asymmetrical Variance Ratio, Common Spatial Pattern (CSP), Ant Colony Optimization (ACO), and a Convolutional Neural Network (CNN) with a Policy Network for effective channel selection. For feature selection, we employ the BorutaPy algorithm, Lasso Regularization, Feature Importance, and Mutual Information to optimize the feature space. Multiple classifiers, including Random Forest, AdaBoost, Gradient Boosting, Support Vector Machine (SVM), and a Stacking classifier, are used to enhance classification performance. The DEAP dataset, which includes emotional responses to various stimuli, is utilized to validate the framework across binary, multi-label, and 3-class classification tasks. Experimental results demonstrate that the proposed channel selection methods significantly improve emotion detection accuracy compared to baseline models, making this approach highly promising for EEG-based emotion recognition applications.