EEG-based emotion recognition in music listening

Area of Science:

• Neuroscience
• Computational Neuroscience
• Affective Computing

Background:

• Ongoing brain activity, recorded via electroencephalograph (EEG), offers insights into the relationship between emotional states and neural dynamics.
• Understanding these links is crucial for developing objective measures of emotional experience.

Purpose of the Study:

• To apply machine learning algorithms for categorizing EEG dynamics based on self-reported emotional states during music listening.
• To develop a framework for optimizing EEG-based emotion recognition by identifying emotion-specific features and evaluating classifier performance.

Main Methods:

• Utilized machine learning, specifically Support Vector Machines (SVM), to classify four emotional states: joy, anger, sadness, and pleasure.
• Systematically identified 30 subject-independent EEG features relevant to emotional processing and explored the use of fewer electrodes.

Main Results:

• Achieved an average classification accuracy of 82.29% +/- 3.06% across 26 subjects for emotion recognition using EEG.
• Identified key features predominantly from frontal and parietal lobe electrodes, aligning with existing literature.
• Demonstrated the feasibility of characterizing EEG dynamics with a reduced number of electrodes.

Conclusions:

• The study successfully established a high accuracy rate for EEG-based emotion recognition using machine learning.
• Identified specific, subject-independent EEG features that can reliably indicate emotional states during music listening.
• Suggests potential for a practical, noninvasive system for assessing emotional states in clinical and real-world settings.