Applying Big Transfer-based classifiers to the DEAP dataset.

Area of Science:

• Neuroscience
• Machine Learning
• Psychology

Background:

• Affective brain-computer interfaces (BCIs) are rapidly advancing.
• Accurate emotion recognition from physiological signals is crucial for psychology and human-computer interaction.
• The DEAP dataset is widely used for emotion classification research.

Purpose of the Study:

• To develop accurate emotion classification models using the DEAP dataset.
• To investigate the effectiveness of heat maps from spectral neurological data for emotion recognition.
• To evaluate Big Transfer neural networks for classifying arousal, valence, and dominance.

Main Methods:

• Generated heat maps from spectral data of neurological signals in the DEAP dataset.
• Addressed class imbalance by discarding data from the majority class.
• Fine-tuned Big Transfer neural networks for binary classification of affective states.

Main Results:

• Achieved over 98% accuracy in classifying arousal, valence, and dominance.
• Attained over 990% balanced accuracy across all three classification tasks.
• Investigated the impact of the data balancing method on classifier performance.

Conclusions:

• The proposed method of using heat maps and a specific data balancing technique is effective for emotion classification.
• Big Transfer neural networks show high performance in affective state recognition.
• This approach offers a promising direction for developing advanced affective BCIs.