Binary classification of multichannel-EEG records based on the ϵ-complexity of continuous vector functions

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Feature selection is critical for accurate electroencephalogram (EEG) classification.
• High-dimensional feature spaces are commonly used but pose challenges.
• Developing a low-dimensional feature space is essential for efficient EEG analysis.

Purpose of the Study:

• To develop a model-free method for binary classification of EEG records.
• To create a low-dimensional feature space for EEG signal analysis.
• To enhance the efficiency and accuracy of EEG classification.

Main Methods:

• The proposed approach extends the theory of ϵ-complexity to vector functions for multichannel EEG records.
• Feature extraction involves estimating ϵ-complexity coefficients of the signal and its finite differences.
• Classification is performed using Random Forest (RF) or Support Vector Machine (SVM) algorithms.

Main Results:

• The Random Forest classifier achieved superior performance, with an out-of-bag accuracy of 85.3%.
• 10-fold cross-validation demonstrated an average accuracy of 84.5% for RF and 81.07% for SVM.
• The method successfully reduced the feature space to four dimensions, outperforming classical spectral feature classification.

Conclusions:

• A novel, model-free method for binary classification of EEG records has been developed.
• The proposed technique effectively reduces feature space dimensionality to four dimensions.
• The results confirm the method's effectiveness and potential for improved EEG signal analysis.