Z-score linear discriminant analysis for EEG based brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Linear Discriminant Analysis (LDA) is a common classification method in Brain-Computer Interfaces (BCI).
• Standard LDA relies on assumptions of Gaussian data distribution and equal covariance matrices, which are often violated in real BCI data.
• Heteroscedastic class distributions are frequently observed in BCI applications, limiting LDA's effectiveness.

Purpose of the Study:

• To propose an enhanced LDA algorithm, Z-score Linear Discriminant Analysis (Z-LDA), to address heteroscedasticity in BCI data.
• To introduce a novel decision boundary definition strategy for improved classification accuracy in BCI.

Main Methods:

• Developed Z-score Linear Discriminant Analysis (Z-LDA) by defining decision boundaries using z-scores.
• Z-LDA incorporates both mean and standard deviation of projected data to adapt the decision boundary.
• Evaluated Z-LDA performance on simulated and two real-world BCI datasets.

Main Results:

• Z-LDA demonstrated significantly higher average classification accuracies compared to conventional LDA.
• The proposed z-score decision boundary strategy effectively handles heteroscedastic class distributions.
• Consistent improvements were observed across simulation and actual BCI datasets.

Conclusions:

• Z-LDA offers a superior alternative to standard LDA for BCI applications with heteroscedastic data.
• The adaptive decision boundary strategy of Z-LDA enhances classification performance in challenging BCI scenarios.
• This method shows promise for improving the reliability and accuracy of BCI systems.