Transferred Subspace Learning Based on Non-negative Matrix Factorization for EEG Signal Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalogram (EEG) signal classification is a key area in neuroscience and machine learning.
• Traditional methods often fail when training and testing datasets originate from different distributions.
• Domain adaptation is crucial for practical EEG analysis where data distributions vary.

Purpose of the Study:

• To propose a novel method for EEG signal classification that addresses domain shift challenges.
• To enhance the utilization of knowledge from training datasets for improved performance on testing datasets.
• To develop a robust approach for EEG classification in real-world applications.

Main Methods:

• A novel Non-negative Matrix Factorization Transfer Learning (NMF-TL) approach is introduced.
• Non-negative Matrix Factorization (NMF) is employed to extract a shared subspace between training and testing datasets.
• The extracted shared subspace is combined with the original feature space for classification.

Main Results:

• The NMF-TL method effectively extracts essential shared information between datasets.
• Combining the shared subspace with the original feature space maximizes signal utilization.
• Extensive experiments on the Bonn EEG dataset validate the proposed method's effectiveness.

Conclusions:

• The NMF-TL method offers a powerful solution for EEG signal classification with domain adaptation.
• This approach improves classification accuracy by effectively bridging domain gaps.
• The findings highlight the potential of NMF-TL in practical, real-world EEG applications.