A comprehensive FastAPI-based backend service for multi-modal stress detection using physiological signals, psychological questionnaires (DASS-21), and voice analysis. The system employs Explainable AI (XAI) techniques to provide interpretable stress predictions.

• Multi-Modal Stress Detection: Combines physiological, psychological, and voice data
• Explainable AI (XAI): SHAP and LIME-based explanations for predictions
• Late Fusion Architecture: Intelligent combination of multiple modalities
• Real-time Processing: Fast inference with optimized feature extraction
• RESTful API: Easy integration with frontend applications
• Docker Support: Containerized deployment ready

1. Physiological Signals (ECG, EDA, EMG, Temperature)
2. Psychological Assessment (DASS-21 Questionnaire)
3. Voice Analysis (Optional audio processing)

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Physiological │    │   DASS-21       │    │   Voice         │
│   Data (CSV)    │    │   Responses     │    │   Probabilities │
└─────────┬───────┘    └─────────┬───────┘    └─────────┬───────┘
          │                      │                      │
          ▼                      ▼                      ▼
┌─────────────────────────────────────────────────────────────────┐
│                    Feature Extraction                           │
│  • Time-domain features (mean, std, skew, kurtosis)            │
│  • Frequency-domain features (power spectral density)          │
│  • Wavelet features (multi-resolution analysis)                │
│  • ECG-specific features (RR intervals, heart rate)            │
└─────────────────────────────────────────────────────────────────┘
          │                      │                      │
          ▼                      ▼                      ▼
┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│ Physiological   │    │ DASS-21         │    │ Voice           │
│ Model           │    │ Model           │    │ Model           │
└─────────┬───────┘    └─────────┬───────┘    └─────────┬───────┘
          │                      │                      │
          └──────────────────────┼──────────────────────┘
                                 ▼
                    ┌─────────────────────────┐
                    │   Late Fusion Model     │
                    │  (PhysioDominantFusion) │
                    └─────────────┬───────────┘
                                  ▼
                    ┌─────────────────────────┐
                    │   XAI Explanations      │
                    │  (SHAP + LIME)          │
                    └─────────────┬───────────┘
                                  ▼
                    ┌─────────────────────────┐
                    │   Stress Prediction     │
                    │  (Low/Medium/High)      │
                    └─────────────────────────┘

• Python 3.10+
• pip
• Docker (optional)

1. Clone the repository

   git clone <repository-url>
   cd Safespace_fastapi

2. Create virtual environment

   python -m venv newenv
   # On Windows
   newenv\Scripts\activate
   # On macOS/Linux
   source newenv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Run the application

   uvicorn main:app --host 0.0.0.0 --port 8080 --reload

1. Build the Docker image

   docker build -t safespace-api .

2. Run the container

   docker run -p 8080:8080 safespace-api

POST /predict

Combines physiological data, DASS-21 responses, and optional voice probabilities to predict stress levels.

• Content-Type: multipart/form-data
• Parameters:
  • physiological_file: CSV file with physiological data
  • dass21_responses: DASS-21 responses (comma-separated or JSON)
  • voice_probabilities: Voice probabilities (optional, comma-separated or JSON)

curl -X POST "http://localhost:8080/predict" \
  -H "Content-Type: multipart/form-data" \
  -F "[email protected]" \
  -F "dass21_responses=1,2,3,1,2,3,1" \
  -F "voice_probabilities=0.2,0.5,0.3"

{
  "prediction": {
    "stress_level": "Medium",
    "confidence": 0.85,
    "probabilities": {
      "low": 0.15,
      "medium": 0.85,
      "high": 0.00
    }
  },
  "explanations": {
    "physiological": {
      "available": true,
      "method": "SHAP",
      "feature_importance": [
        {
          "feature": "ECG_mean_rr",
          "importance": 0.25,
          "abs_importance": 0.25
        }
      ],
      "summary": "ECG heart rate variability is the most important factor..."
    },
    "dass21": {
      "available": true,
      "method": "SHAP",
      "feature_importance": [
        {
          "feature": "DASS21_Q3_positive_feelings",
          "importance": -0.30,
          "abs_importance": 0.30
        }
      ],
      "summary": "Positive feelings score significantly influences the prediction..."
    },
    "fusion": {
      "available": true,
      "method": "Late Fusion",
      "modality_contributions": {
        "physiological": 0.60,
        "dass21": 0.25,
        "voice": 0.15
      },
      "summary": "Physiological signals contribute most to the final prediction..."
    }
  },
  "processing_info": {
    "windows_processed": 10,
    "features_extracted": 180,
    "processing_time_ms": 245
  }
}

Safespace_fastapi/
├── main.py                 # Main FastAPI application
├── latefusion_final.py     # Late fusion model implementation
├── predict_wesad.py        # WESAD dataset prediction utilities
├── predict_physiological.py # Physiological data processing
├── run_wesad_prediction.py # WESAD prediction runner
├── v1.py                   # Version 1 API endpoints
├── requirements.txt        # Python dependencies
├── Dockerfile             # Docker configuration
├── scaler.pkl             # Feature scaler
├── models/                # Trained model files
│   ├── fusion_model.pkl
│   ├── lateFusion.pkl
│   ├── regularized_global_model.pkl
│   ├── stacking_classifier_model.pkl
│   ├── scaler.pkl
│   └── Voice.h5
└── newenv/                # Virtual environment (ignored by git)

CFG = {
    "orig_fs": 700,        # Original sampling frequency
    "fs": 100,             # Target sampling frequency
    "window_sec": 10,      # Window size in seconds
    "stride_sec": 5,       # Stride between windows
    "sensors": ["ECG", "EDA", "EMG", "Temp"]  # Supported sensors
}

mod_weights = {
    'phys': 0.60,    # Physiological signals (60%)
    'text': 0.25,    # DASS-21 responses (25%)
    'voice': 0.15    # Voice analysis (15%)
}

Expected columns for each sensor:

• ECG: Raw ECG signal values
• EDA: Electrodermal activity values
• EMG: Electromyography values
• Temp: Temperature values

7 questions with responses 0-3:

• 0: Did not apply to me at all
• 1: Applied to me to some degree, or some of the time
• 2: Applied to me to a considerable degree, or a good part of the time
• 3: Applied to me very much, or most of the time

Three probability values for stress levels:

• [low_prob, medium_prob, high_prob]

1. Time-domain Features:

   • Statistical measures (mean, std, variance, skewness, kurtosis)
   • Range measures (min, max, peak-to-peak)
   • Percentiles (25th, 75th, median)

2. Frequency-domain Features:

   • Power spectral density in different bands
   • Frequency statistics (mean, std, peak frequency)

3. Wavelet Features:

   • Multi-resolution analysis using PyWavelets
   • Detail coefficients (d1-d4) and approximation coefficients (a4)

4. ECG-specific Features:

   • RR intervals and heart rate variability
   • Heart rate statistics

• SHAP (SHapley Additive exPlanations): For feature importance analysis
• LIME (Local Interpretable Model-agnostic Explanations): For local explanations
• Permutation Importance: For model-agnostic feature ranking

• Inference Time: ~250ms per prediction
• Feature Extraction: ~180 features per window
• Window Processing: 10-second windows with 5-second stride
• Memory Usage: ~500MB (including models)

Once the server is running, visit:

• Interactive API Docs: http://localhost:8080/docs
• ReDoc Documentation: http://localhost:8080/redoc
• OpenAPI Schema: http://localhost:8080/openapi.json

# Sample DASS-21 responses
dass21_responses = "1,2,3,1,2,3,1"

# Sample voice probabilities
voice_probabilities = "0.2,0.5,0.3"

# Test with sample data
curl -X POST "http://localhost:8080/predict" \
  -H "Content-Type: multipart/form-data" \
  -F "physiological_file=@sample_data.csv" \
  -F "dass21_responses=1,2,3,1,2,3,1"

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.

• SafeSpace Team - Initial work

• WESAD dataset for physiological data
• DASS-21 questionnaire for psychological assessment
• SHAP and LIME libraries for explainable AI
• FastAPI for the web framework

For support and questions:

• Create an issue in the repository
• Contact the development team
• Check the API documentation at /docs

Note: This API is designed for research and educational purposes. For clinical applications, additional validation and medical certification may be required.