Statistically significant features improve binary and multiple Motor Imagery task predictions from EEGs

Area of Science:

• Neuroscience and Biomedical Engineering
• Focus on Brain-Computer Interface (BCI) systems
• Utilizing electroencephalogram (EEG) signals for human-computer interaction

Background:

• Motor Imagery (MI) tasks are crucial for BCI systems, enabling communication and device control for individuals with paralysis.
• EEG signal analysis for MI-BCI faces challenges due to the non-linear and non-stationary nature of EEG data.
• Effective feature extraction and classification are essential for accurate MI-BCI system design.

Purpose of the Study:

• To investigate the impact of statistical significance-based feature selection on the classification accuracy of Motor Imagery EEG signals.
• To evaluate the effectiveness of various time-domain, frequency-domain, time-frequency domain, and non-linear features.
• To compare classification performance using the full feature set versus a reduced set of statistically significant features.

Main Methods:

• Extracted 1,364 features from 22-channel EEG data, including time-domain, Fourier transform, Wavelet transform, and Poincare plot parameters.
• Employed independent t-test for binary and ANOVA for multi-class classification to identify statistically significant features.
• Classified data using 6-7 different algorithms, evaluated with five-fold cross-validation and repeated 10 times for reliability.

Main Results:

• The Ensemble Subspace Discriminant classifier achieved maximum accuracies of 61.86% for two-class and 47.36% for four-class MI tasks.
• Classification performance improved significantly when using only statistically significant features compared to the entire feature set.
• The study demonstrated that statistical feature selection leads to higher classifier performance with fewer components.

Conclusions:

• Statistical significance-based feature selection is an effective approach to enhance MI-BCI classification accuracy.
• Non-linear parameters offer a valuable alternative to commonly used features in MI-BCI.
• The proposed method facilitates the prediction of multiple Motor Imagery tasks, improving BCI system utility.