EEG Word Classification

This project implements EEG-based imagined speech classification using the FEIS dataset, mapped to the Emotiv Epoc X headset configuration (14 channels, 128 Hz).

The pipeline includes preprocessing, multiple deep learning models (CNN, TCN, RNN), and evaluation under Leave-One-Subject-Out (LOSO) cross-validation.

---

📂 Project Structure

EEG-EpocX-Models/ │── README.md # Project documentation │── requirements.txt # Dependencies │── config.yaml # Global configs (default run) │ ├── data/ │ ├── raw/ # Original FEIS .mat files (64ch, 512Hz) │ ├── preprocessed/ # Downsampled + EpocX channels (14ch, 128Hz) │ └── splits/ # Train/validation/test subject splits │ ├── notebooks/ # Data exploration and visualization ├── src/ # Source code │ ├── data/ # Dataset + preprocessing scripts │ ├── models/ # Model definitions (CNN, TCN, RNN, MLP) │ └── training/ # Train/eval loops, utilities │ └── experiments/ # Model-specific YAML configs ├── exp1_cnn.yaml ├── exp2_tcn.yaml └── exp3_rnn.yaml

yaml Copy code

---

⚙️ Setup

1. Create a virtual environment (recommended):

python -m venv .venv
source .venv/bin/activate   # macOS/Linux
.venv\Scripts\activate      # Windows
Install dependencies:

bash
Copy code
pip install -r requirements.txt
Prepare dataset:

Download the FEIS dataset from OSF or Kaggle.

Place subject .mat files in:

bash
Copy code
data/raw/FEIS/train/
data/raw/FEIS/validation/
🧹 Preprocessing
Convert FEIS (64ch, 512Hz) → EpocX (14ch, 128Hz):

bash
Copy code
python -m src.data.preprocess --raw_dir data/raw/FEIS/train --out_dir data/preprocessed/train
python -m src.data.preprocess --raw_dir data/raw/FEIS/validation --out_dir data/preprocessed/validation
This will:

Downsample from 512 Hz → 128 Hz

Select EpocX channels:

bash
Copy code
['AF3','F7','F3','FC5','T7','P7','O1',
 'O2','P8','T8','FC6','F4','F8','AF4']
Normalize each trial (zero mean, unit variance)

Save .npz files to data/preprocessed/

🏋️ Training
The project supports multiple architectures:

EEGNet-style CNN → experiments/exp1_cnn.yaml

Temporal Convolutional Network (TCN) → experiments/exp2_tcn.yaml

BiLSTM with Attention → experiments/exp3_rnn.yaml

Baseline MLP → src/models/baseline_mlp.py

Train with:

bash
Copy code
python -m src.training.train --config experiments/exp1_cnn.yaml
Features:

Leave-One-Subject-Out cross-validation (15 folds)

Early stopping on validation accuracy

Checkpointing of best model

TensorBoard logging

📊 Results
Outputs are saved in results/:

results/logs/ → TensorBoard logs

results/checkpoints/ → Model weights

results/figures/ → Confusion matrices, plots

View training progress:

bash
Copy code
tensorboard --logdir results/logs