Skip to content

Application and Usage

Mental Health Monitoring Prototype

This application demonstrates how embedding-based NLP models and SVM classifiers can be integrated into a functional longitudinal monitoring system.

The prototype accepts Arabic text input, predicts depression and anxiety severity levels, and generates structured alerts based on temporal trends.

This is a research prototype intended for experimentation and demonstration purposes only.

System Workflow

  • 1. Text Input

The user provides an Arabic narrative entry along with an optional participant ID and timestamp.

  • 2. Embedding Generation

The text is converted into a 768-dimensional vector representation using a multilingual embedding model.

  • 3. Severity Prediction

Two independent SVM classifiers predict: - Depression severity (0 to 3) - Anxiety severity (0 to 3)

  • 4. Alert Evaluation

Recent historical scores are analyzed to detect: - Sustained high severity - Worsening trends - Sudden spikes

  • 5. Output Response

The system returns severity scores and any triggered alert categories.

Running the Application

Local Execution

  1. Clone the repository.
  2. Install dependencies:
pip install -r requirements.txt
````

3. Start the application server:

```bash
python app.py
  1. Open your browser at:
http://localhost:8000

Example Usage

Input

{
  "text": "احس اني متعبة وما عندي طاقة",
  "participant_id": "P023",
  "date": "2026-02-22"
}

Output

{
  "depression_score": 2,
  "anxiety_score": 1,
  "alerts": []
}

Alert Logic Overview

  • High Score Alert

Triggered when the average score over the last three entries is 2 or higher.

  • Worsening Trend Alert

Triggered when the average increase exceeds 0.5 across three consecutive entries.

  • Sudden Spike Alert

Triggered when the score increases by 2 points between consecutive entries.

Intended Use

The application is designed as a structured research demonstration of longitudinal mental health modeling.

It is not:

  • A diagnostic tool
  • A medical recommendation system
  • A replacement for clinical evaluation

It is intended for:

  • Academic experimentation
  • AI methodology demonstration
  • Longitudinal modeling research

Deployment Considerations

If deployed beyond local testing, the following safeguards are recommended:

  • Authentication and access control
  • Secure logging with sensitive data filtering
  • Rate limiting
  • Human-in-the-loop review for high-risk alerts
  • Clear user disclaimers and consent mechanisms

Technical Stack

Component Technology
Embedding Model Multilingual embedding model (768-dim vectors)
Classifier Support Vector Machine with RBF kernel
Backend Python-based API service
Data Synthetic longitudinal Arabic dataset

Limitations

  • Performance is evaluated on synthetic data.
  • Real-world deployment requires clinical validation.
  • Alert thresholds are rule-based and may require calibration.