NeuroSense: A Novel EEG Dataset Utilizing Low-Cost, Sparse Electrode Devices for Emotion Exploration

Emotion recognition is crucial in affective computing, aiming to bridge the gap between human emotional states and computer understanding. This study presents NeuroSense, a novel electroencephalography (EEG) dataset utilizing low-cost, sparse electrode devices for emotion exploration. Our dataset comprises EEG signals collected with the portable 4-electrodes device Muse 2 from 30 participants who, thanks to a neurofeedback setting, watch 40 music videos and assess their emotional responses. These assessments use standardized scales gauging arousal, valence, and dominance. Additionally, participants rate their liking for and familiarity with the videos. We develop a comprehensive preprocessing pipeline and employ machine learning algorithms to translate EEG data into meaningful insights about emotional states. We verify the performance of machine learning (ML) models using the NeuroSense dataset. Despite utilizing just 4 electrodes, our models achieve an average accuracy ranging from 75% to 80% across the four quadrants of the dimensional model of emotions. We perform statistical analyses to assess the reliability of the self-reported labels and the classification performance for each participant, identifying potential discrepancies and their implications. We also compare our results with those obtained using other public EEG datasets, highlighting the advantages and limitations of sparse electrode setups in emotion recognition. Our results demonstrate the potential of low-cost EEG devices in emotion recognition, highlighting the effectiveness of ML models in capturing the dynamic nature of emotions. The NeuroSense dataset is publicly available, inviting further research and application in human-computer interaction, mental health monitoring, and beyond.