Motor Imagery Classification
Binary and multi-class decoding of imagined limb movements from multi-channel EEG recordings.
Research
EEG-based motor imagery classification at Universidad de Antioquia: bridging signal processing, classical ML, and software engineering.
Overview
My master's research tackles motor imagery classification from EEG: distinguishing imagined left-hand vs. right-hand movements from noisy scalp recordings. This is a foundational building block for non-invasive brain-computer interfaces that could assist motor rehabilitation or device control.
The work combines MNE-based preprocessing, Filter Bank Common Spatial Patterns (FBCSP), and linear classifiers evaluated with leave-one-subject-out cross-validation. I document every step so experiments can be reproduced and compared fairly.
Research Topics
FBCSP Feature Extraction
Spatial filtering across frequency bands to maximize class separability before classification.
Subject-Disjoint Evaluation
Leave-one-subject-out cross-validation, confusion matrices, and per-participant performance analysis.
Reproducible Pipelines
Python workflows from raw .edf files to trained models with versioned preprocessing and logging.
Methods & Tools
- EEG acquisition and band-pass / notch filtering
- FBCSP and CSP spatial feature extraction
- LDA, SVM, and k-means clustering classifiers
- scikit-learn, pandas, NumPy, and MNE-Python
- Leave-one-subject-out cross-validation
- Confusion matrices and statistical performance reporting
Research Goals
Short term: refine the motor imagery pipeline, compare FBCSP configurations, and publish reproducible benchmarks under consistent evaluation protocols.
Long term: connect research prototypes with production-grade software that makes neurotechnology more accessible: interpretable models, clean APIs, and tools researchers can actually deploy.