Emotion classification based on gamma-band EEG.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Emotion recognition is a key area in affective computing.
• Electroencephalography (EEG) offers a non-invasive method for brain activity monitoring.
• Distinguishing between basic emotions like happiness and sadness is crucial for human-computer interaction.

Purpose of the Study:

• To develop and validate a method for classifying happiness and sadness from EEG signals.
• To identify an optimal frequency band for emotion classification using EEG.
• To evaluate the impact of trial length on classification accuracy.

Main Methods:

• Utilized electroencephalography (EEG) signals recorded during the presentation of facial expressions (smiles and cries).
• Employed a frequency band searching method to determine the optimal EEG signal filter band.
• Applied Common Spatial Patterns (CSP) for feature extraction and a linear Support Vector Machine (SVM) for classification.
• Investigated classification performance using trial lengths of 3 seconds and 1 second.

Main Results:

• Achieved high classification accuracies for both happiness and sadness: 93.5% +/- 6.7% for 3s-trials and 93.0% +/- 6.2% for 1s-trials.
• Identified the gamma frequency band (approximately 30-100 Hz) as optimal for EEG-based emotion classification.
• Demonstrated that accurate emotion classification is achievable even with short trial durations (1s).

Conclusions:

• EEG signals, particularly within the gamma band, are effective for classifying emotions like happiness and sadness.
• The proposed method combining frequency band searching, CSP, and linear-SVM provides a robust approach to emotion recognition.
• The findings suggest potential for real-time emotion detection systems using EEG.