EEG QA + ML Pipeline (DEAP Dataset)

This project simulates a QA-tested machine learning pipeline for EEG classification using real signal structure from the DEAP dataset. It includes feature extraction, model training, evaluation, and CI/CD automation — showcasing testing best practices and explainability.

---

Dataset Used

• Source: DEAP - Dataset for Emotion Analysis using Physiological signals
• Input: EEG features extracted from 32 scalp channels (CSV format)
• Labels: Simulated binary arousal labels ("high_arousal", "low_arousal")
• Note: Label simulation mirrors DEAP’s 1–9 arousal ratings using a thresholded normal distribution

---

Pipeline Overview

• mock_eeg_signals.csv: Raw EEG features (real signals, no labels)
• signal_filter.py: Stubbed pre-processing (bandpass placeholder)
• feature_extraction.py: Extracts delta, theta, alpha, beta, gamma band powers
• train_classifier.py: Simulates labels, trains RandomForest, outputs metrics
• predict.py: Generates predictions using saved model
• metrics.txt: Accuracy, precision, recall, confusion matrix
• top_features.csv: Feature importance scores
• top_10_features_plot.png: Horizontal bar chart of most predictive features
• confusion_matrix.png: Confusion matrix visualization

---

Tools Used

• Python: pandas, numpy, scikit-learn
• Visualization: matplotlib, seaborn
• Testing: pytest
• CI/CD: GitHub Actions
• Explainability: Feature importance analysis

---

QA & Test Coverage

• Unit tests for filtering and feature extraction
• End-to-end test for the entire pipeline
• CI pipeline runs full test suite on every push (.github/workflows/ci.yml)

---

Running the Project

pip install -r requirements.txt

# Run the full pipeline
python3 processing/signal_filter.py
python3 processing/feature_extraction.py
python3 model/train_classifier.py
python3 model/predict.py

# Run tests
pytest

---

Results & Insights

• Accuracy: ~48% with structured simulated labels
• Balanced classification of arousal states
• Most important EEG features: Cz_gamma, Fp1_alpha, Cz_alpha, Cz_delta, Fp1_beta
• Visuals saved: confusion_matrix.png, top_10_features_plot.png

---

What I Learned

• Building a modular QA-friendly ML pipeline
• Simulating biologically inspired labels
• Implementing test coverage across an ML stack
• Automating with GitHub Actions
• Explaining model behavior using feature importance

---

Future Improvements

• Load real DEAP arousal/valence ratings
• Add true bandpass filtering (e.g., Butterworth)
• Improve feature engineering
• Explore deep learning alternatives
• Wrap in a Jupyter Notebook for interactive use

---

Data format based on DEAP EEG dataset.