Classificação de Valência emocional em sinais de eletroencefalograma por meio de métodos de aprendizado de máquina

Abstract

Emotion recognition using electroencephalogram (EEG) signals has gained increasing attention due to its potential applications in Affective Computing and Brain-Computer Interfaces. This study proposes an approach for emotional valence classification in EEG signals using machine learning methods. The public Imagined Emotion Study dataset (OpenNeuro ds003004), composed of recordings from 34 participants and 511 emotional events labeled as positive or negative valence, was used. After signal preprocessing, temporal, spectral, and differential entropy features were extracted. Different feature sets, temporal window lengths, and feature selection strategies were evaluated using the K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest(RF) classifiers. Model evaluation was performed using cross-validation, ensuring complete separation of participants between training and testing sets. The results showed that the combination of temporal, spectral, and differential entropy features, together with feature selection and hyperparameter optimization, improved classification performance. The best overall performance was achieved by the SVM classifier, reaching an accuracy of 66.29% and a balanced accuracy of 65.98%, while RF obtained the highest F1-score (69.24%). The findings indicate that the proposed methodology is capable of identifying discriminative patterns associated with emotional valence in EEG signals, achieving results comparable to those reported in the literature for inter-subject validation scenarios.