EEG Motor-Imagery Brain-Computer Interface Using CSP and Machine Learning

A brain-computer interface decides from 22-channel EEG whether a person imagines moving the left or the right hand (and, in the extension, both feet or the tongue) - without any muscle movement. The project uses the real BCI Competition IV data set 2a (9 subjects, 250 Hz, two sessions on different days) and the classic ECE chain: an 8-30 Hz Butterworth band-pass for the mu and beta rhythms, a 2-s window after the cue, Common Spatial Patterns (3 + 3 filters), log-variance features and LDA. It is compared with filter-bank CSP + SVM, a Riemannian tangent-space classifier and the compact CNN EEGNet. Every model is trained on session T and tested once on session E (another day): left vs right CSP + LDA reaches 78.6 +- 16.5 % (kappa 0.573) with a spread from 55.6 % to 97.2 % between subjects; in 4 classes the tangent-space method reaches kappa 0.564, close to the competition winner (0.57). MNE, MOABB, GNU Octave and EEGLAB were run as cross-checks; a complete MATLAB version, an online cursor demo and a web application with five databases are included.

Category: Electronics and Communication Engineering (Biomedical Signal Processing, EEG, Brain-Computer Interface, Common Spatial Patterns, Filter-Bank CSP, Riemannian Geometry, EEGNet, MNE-Python, MOABB, GNU Octave, EEGLAB, MATLAB, Python)

Electronics & Communication Engineering Project E07 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.