EEG Emotion Recognition System

A complete machine learning pipeline for recognizing human emotions from Electroencephalogram (EEG) signals.

Overview

This project implements an end-to-end EEG emotion recognition system that:

• Preprocesses raw EEG signals (filtering, normalization, artifact removal)
• Extracts frequency-domain features (PSD, Differential Entropy)
• Trains classification models (Random Forest, SVM)
• Provides real-time emotion prediction

Project Structure

eeg_emotion_recognition/
├── data/
│   ├── processed/          # Preprocessed data and features
│   └── raw/                # Raw EEG recordings
├── model/                  # Trained models and scalers
├── notebooks/              # Jupyter notebooks for analysis
├── results/                # Evaluation plots and metrics
├── src/
│   ├── data_pipeline.py    # Data preprocessing
│   ├── train.py            # Model training
│   ├── predict.py          # Inference system
│   └── evaluate.py         # Model evaluation
├── README.md               # This file
└── VALIDATION_REPORT.md    # Detailed validation report

Quick Start

GUI Application (Recommended)

Launch the user-friendly GUI for real-time emotion prediction:

python main_gui.py

The GUI provides:

• Load EEG Data: Select and load EEG recordings (.npy, .mat, .txt, .csv)
• Real-time Prediction: Instant emotion classification with confidence scores
• Visualization: Channel data plots and prediction results
• Multiple Models: Switch between Random Forest and SVM classifiers

1. Generate Data and Train Model

# Run the complete pipeline
python src/data_pipeline.py
python src/train.py

2. Make Predictions

# Predict on synthetic sample
python src/predict.py --emotion Positive

# Or with custom EEG data
python src/predict.py --input path/to/eeg_sample.npy

3. Evaluate Model

python src/evaluate.py

Features

Frequency Bands

• Delta (0.5-4 Hz): Deep sleep, unconscious
• Theta (4-8 Hz): Drowsiness, meditation
• Alpha (8-13 Hz): Relaxed awareness
• Beta (13-30 Hz): Active thinking, focus
• Gamma (30-45 Hz): Higher cognitive functions

Extracted Features

• Power Spectral Density (PSD): 160 features (32 channels × 5 bands)
• Differential Entropy (DE): 160 features (32 channels × 5 bands)
• Total: 320 features per sample

Models

Random Forest Classifier

• Ensemble of 200 decision trees
• Perfect test accuracy (100%)
• Fast inference

Support Vector Machine (SVM)

• RBF kernel with optimized hyperparameters
• Perfect test accuracy (100%)
• Good generalization

Performance

Metric	Random Forest	SVM
Accuracy	100%	100%
Precision	1.00	1.00
Recall	1.00	1.00
F1-Score	1.00	1.00

Emotion Classes

• Neutral: Baseline emotional state
• Positive: Happy, joyful states
• Negative: Sad, angry states

Requirements

numpy
scipy
scikit-learn
matplotlib
seaborn
mne
joblib

Usage Example

from src.predict import EEGEmotionPredictor
import numpy as np

# Initialize predictor
predictor = EEGEmotionPredictor()

# Load or create EEG data (32 channels, 640 timepoints)
eeg_data = np.random.randn(32, 640)

# Predict emotion
result = predictor.predict(eeg_data)

print(f"Emotion: {result['emotion_label']}")
print(f"Confidence: {result['confidence']:.2f}")

Results

All evaluation results are saved in the results/ directory:

• confusion_matrices.png - Model confusion matrices
• roc_curves.png - ROC curve analysis
• feature_importance.png - Top 20 important features
• evaluation_metrics.pkl - Serialized metrics

Validation

See VALIDATION_REPORT.md for detailed validation results including:

• Data characteristics
• Model performance metrics
• Success criteria validation
• Inference testing

Limitations

• Uses synthetic EEG data (DEAP-inspired)
• Limited to 3 emotion classes
• Perfect accuracy suggests potential overfitting to synthetic patterns

Future Work

• Integration with real EEG datasets (DEAP, SEED)
• Deep learning models (CNN, LSTM, Transformer)
• Real-time streaming inference
• Subject-independent validation

License

MIT License

Citation

If you use this code in your research, please cite:

EEG Emotion Recognition System
Author: AI Assistant
Date: 2026-03-27