A TrAdaBoost Method for Detecting Multiple Subjects' N200 and P300 Potentials Based on Cross-Validation and an...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Subject-specific classifiers in brain-computer interfaces (BCIs) traditionally require extensive data per user.
• This data collection leads to user fatigue and increased training burden.

Purpose of the Study:

• To propose and evaluate a novel training method, CV-T-TAB, for BCIs.
• To reduce the amount of data needed for training subject-specific classifiers.

Main Methods:

• The proposed Cross-Validation and Adaptive Threshold TrAdaBoost (CV-T-TAB) method combines classifiers from multiple subjects.
• Cross-validation artificially extends the training data for a new subject.
• An adaptive threshold selects optimal classifier combinations.

Main Results:

• CV-T-TAB demonstrated superior performance compared to five traditional methods in training support vector machines.
• High accuracy was maintained even with a reduction in training data to one-third.
• Performance metrics included accuracy, information transfer rate, AUC, recall, and precision.

Conclusions:

• CV-T-TAB effectively improves subject-specific classifier performance using limited data.
• The method leverages classifiers from multiple subjects to reduce overall training costs.
• This approach offers a more efficient and less burdensome training paradigm for BCIs.