Emotion Recognition From Multi-Channel EEG via Deep Forest.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Deep neural networks (DNNs) show promise in electroencephalography (EEG)-based emotion recognition but require extensive data and hyperparameter tuning.
• Traditional algorithms often lack the efficiency and accuracy needed for complex emotion recognition tasks.

Purpose of the Study:

• To propose a novel deep forest method for multi-channel EEG-based emotion recognition that overcomes DNN limitations.
• To enhance the accuracy and reduce the complexity of emotion recognition from EEG signals.

Main Methods:

• Preprocessed raw EEG signals using baseline removal to mitigate baseline signal effects.
• Constructed 2D frame sequences from multi-channel EEG, considering spatial relationships.
• Employed a deep forest classification model to extract spatial-temporal information for emotion classification.

Main Results:

• Achieved high average accuracies on the DEAP database: 97.69% for valence and 97.53% for arousal.
• Attained significant average accuracies on the DREAMER database: 89.03% (valence), 90.41% (arousal), and 89.89% (dominance).
• Demonstrated superior performance compared to state-of-the-art methods in EEG-based emotion recognition.

Conclusions:

• The proposed deep forest method effectively recognizes emotions from EEG signals with reduced complexity and data requirements.
• This approach eliminates the need for manual feature extraction and is less sensitive to hyperparameter settings.
• The method shows high potential for practical applications in affective computing and brain-computer interfaces.