Improving EEG-Based Emotion Classification Using Conditional Transfer Learning.

Area of Science:

• Neuroscience
• Machine Learning
• Affective Computing

Background:

• Accurate electroencephalogram (EEG)-based emotion classification requires extensive individual calibration data.
• Transfer learning (TL) offers a solution by leveraging existing data, but blind transfer can degrade performance.
• Individual differences in EEG signals necessitate tailored approaches for effective emotion recognition.

Purpose of the Study:

• To introduce a conditional transfer learning (cTL) framework to improve subject-specific emotion classification performance.
• To enable positive transfer by selectively leveraging data from individuals with comparable EEG features.
• To reduce the labor and time associated with extensive ecological calibration data collection.

Main Methods:

• Developed a cTL framework that assesses individual transferability for positive transfer.
• Implemented selective data leveraging from individuals with similar EEG signatures.
• Evaluated the cTL framework's performance on emotion valence and arousal classification tasks.

Main Results:

• The cTL framework identified transferable individuals who benefited from others' data for emotion classification.
• Transferable individuals achieved maximal TL improvements by leveraging data from those with similar EEG signatures.
• cTL improved overall classification performance by approximately 15% for valence and 12% for arousal.

Conclusions:

• The proposed cTL framework is feasible for enhancing individual emotion classification performance using existing data repositories.
• cTL facilitates robust emotion classification models with reduced subject-specific labeled data requirements.
• This approach shows promise for developing real-life affective brain-computer interfaces (ABCIs).