Project Overview
The system decodes motor intent from EEG data, trains a CSP + LDA classifier, and streams live predictions with confidence-weighted feedback.
EEG Motor Action Classifier
A real-time brain-computer interface pipeline that classifies grab vs throw motor actions from EEG signals with offline training and live prediction.
Tech Stack
- → MNE-Python and pyxdf
- → CSP + LDA classifier
- → Lab Streaming Layer (LSL)
- → scikit-learn and calibration
What It Does
- ✓ Trains on XDF EEG recordings with preprocessing and epoch extraction
- ✓ Uses nested cross-validation and hyperparameter tuning for evaluation
- ✓ Streams live EEG and marker data to produce real-time predictions
- ✓ Logs predictions and provides confidence-weighted visual feedback
Offline Training
Training covers EEG preprocessing, epoching around grab and throw markers, CSP feature extraction, calibrated LDA fitting, and model artifact export.
Online Prediction
The online script connects to LSL streams, maintains a rolling EEG buffer, and emits predictions during the analysis window for live feedback.