EEG Mental Attention State Classification

Investigating classical machine learning techniques for classifying human mental attention states (Focused vs. Unfocused) from EEG data, targeting real-time Brain-Computer Interface (BCI) applications using the NeuroPawn 125 Hz headset.

Dataset

XJTU-EEG MEMA (Multi-label EEG for Mental Attention states)

• Source: https://github.com/XJTU-EEG/MEMA
• 20 subjects, 12 trials each, 1,060 minutes total
• Raw data from For_graph/ directory (pre-segmented .txt files)
• 7 channels selected: F3, Fz, F4, C3, Cz, C4, Pz

How to Run

Scripts should be run in order from the project root directory:

# Step 1: Process raw .txt files -> cleaned .npz
python data_processing.py

# Step 2: Extract features from processed EEG data
python feature_extraction.py

# Step 3: Train and evaluate ML models
python model_training.py

# Step 4: Generate visualizations for the report
python visualization.py

Required Libraries

All are standard Anaconda packages:

• numpy, pandas, scipy, scikit-learn, matplotlib, seaborn

Install with: pip install -r requirements.txt

Project Structure

353 project/
data_processing.py       # Raw EEG → cleaned, downsampled windows
feature_extraction.py    # Windows → feature matrices
model_training.py        # Train SVM, KNN, RF, DT, GBoosting
visualization.py         # Generate report figures
results/
   figures/             # Output plots
   metrics/             # CSV result tables
processed_data/          # Intermediate .npz files (generated)
MEMA Dataset/            # Raw data (not included, see above)
Papers/                  # Reference papers
requirements.txt
README.md

Methods

• Preprocessing: Bandpass filter (0.5-45 Hz), downsample 500→125 Hz, 1s windows
• Features: Band powers (delta/theta/alpha/beta/gamma), band ratios, statistical features (84 total)
• Feature Selection: ANOVA, Feature Importance, Linear Correlation, PCA
• Models: SVM, KNN, Random Forest, Decision Tree, Gradient Boosting
• Evaluation: Accuracy, Precision, Recall, F1, Confusion Matrix, 5-fold CV

References

1. Aci et al. (2019) - SVM-based EEG attention classification
2. Wang & Kim (2024) - KNN with feature importance for brain attention detection
3. Liu et al. (2025) - MEMA dataset paper