Improved Transductive Support Vector Machine for a Small Labelled Set in Motor Imagery-Based Brain-Computer...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery-based brain-computer interfaces (BCIs) face challenges due to long calibration times.
• Efficient training methods are crucial for BCI development and usability.

Purpose of the Study:

• To reduce the calibration time for motor imagery-based BCIs.
• To improve the performance of BCIs using limited labeled and large unlabeled datasets.

Main Methods:

• Developed an improved transductive support vector machine (ITSVM) incorporating common spatial patterns (CSP) and geometric features.
• Utilized the concave-convex procedure (CCCP) with a novel balancing constraint to handle unlabeled data distribution.
• Proposed an improved self-training TSVM (IST-TSVM) for iterative feature extraction and classification.

Main Results:

• The proposed IST-TSVM achieved average accuracies of 63.25% and 69.43% on BCI competition datasets IV-a and IV-IIa, respectively.
• Demonstrated superior performance compared to existing algorithms across variable labeled set sizes and distributions.
• Achieved high accuracy with a minimal labeled dataset (4 positive, 16 negative samples).

Conclusions:

• The developed ITSVM and IST-TSVM algorithms effectively reduce BCI calibration time.
• These methods offer a viable alternative for training motor imagery-based BCIs with limited labeled data.
• The findings pave the way for more accessible and efficient BCI applications.