Multi-Feature Input Deep Forest for EEG-Based Emotion Recognition.

Area of Science:

• Affective computing and human-computer interaction.
• Neuroscience and signal processing for emotion recognition.

Background:

• Electroencephalogram (EEG) signals offer a viable, non-intrusive method for emotion recognition.
• High dimensionality and emotional diversity in EEG data pose challenges for accurate feature extraction and pattern recognition.

Purpose of the Study:

• To propose and evaluate a Multi-Feature Deep Forest (MFDF) model for enhanced EEG-based emotion classification.
• To investigate the efficacy of combining Power Spectral Density (PSD) and Differential Entropy (DE) features from various EEG frequency bands.

Main Methods:

• EEG signals were segmented into frequency bands, and PSD and DE features were extracted.
• A deep forest classifier was trained using these multi-features for a five-class emotion model (neutral, angry, sad, happy, pleasant).
• Performance was evaluated on the public DEAP dataset and compared against K Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM).

Main Results:

• The MFDF model achieved an average accuracy of 71.05%, outperforming RF (3.40% higher), KNN (8.54% higher), and SVM (19.53% higher).
• Using dimension-reduced features yielded 51.30% accuracy, while raw EEG signals resulted in only 26.71% accuracy.
• The MFDF model demonstrated superior performance in classifying five distinct emotions.

Conclusions:

• The proposed MFDF model offers a significant advancement in EEG-based emotion classification.
• Effective feature extraction from EEG frequency bands is crucial for improving emotion recognition accuracy.
• This method provides a robust framework for developing more sophisticated affective computing systems.