Robust Feature Engineering Techniques for Designing Efficient Motor Imagery-Based Systems Using Multidomain Electroencephalography

Gaussian-kernel SVM on the MILimbEEG dataset classifies motor imagery from 16-channel EEG with 95.48% testing accuracy after MRMR reduces 416 multidomain features to four.

• Why it matters: EEG’s high dimensionality and signal complexity hinder accurate motor imagery classification for BCI use in neurorehabilitation, motivating more efficient feature engineering and selection.
• What they did: Time-, frequency-, and wavelet-based features were extracted from 16-channel EEG to form 416 features, then MRMR selected the four most informative and minimally redundant features for SVM, KNN, Naive Bayes, and decision tree models.
• The result: The four-feature MRMR representation maintained strong performance, with motor activities reaching 92.50% accuracy and motor imagery achieving precision 86.97%, recall 88.33%, and F1-score 87.04% under 10-fold cross-validation.

AI summary of the abstract · may contain errors · not medical advice · read the paper.