Medical Insurance Cost Predictor 🏥💰

A machine learning web application that predicts medical insurance costs based on demographic and health factors. The application features a modern, interactive frontend with a Flask API backend serving a trained neural network model.

✨ Features

• Interactive Web Interface: Modern, responsive UI with smooth animations
• Real-time Predictions: Get instant insurance cost estimates
• RESTful API: Flask backend with JSON endpoints
• Dockerized Deployment: Easy setup and deployment with Docker
• Machine Learning Model: Neural network trained on insurance data with log transformation

🚀 Demo

Enter your details and get an instant insurance cost prediction

📊 Model Details

• Algorithm: Neural Network
• Features: Age, Sex, BMI, Children, Smoker Status, Region
• Target: Medical Insurance Charges (log10 transformed)
• Framework: scikit-learn

Input Features:

• Age: 18-100 years
• Sex: Male/Female
• BMI: Body Mass Index (15-50)
• Children: Number of dependents (0-10)
• Smoker: Yes/No
• Region: Northeast, Northwest, Southeast, Southwest (US)

🏗️ Architecture

┌─────────────────┐    HTTP/JSON    ┌─────────────────┐
│   Frontend      │ ──────────────→ │   Backend       │
│   (HTML/JS)     │                 │   (Flask API)   │
│   Port: 3000    │ ←────────────── │   Port: 5010    │
└─────────────────┘                 └─────────────────┘
                                            │
                                            ▼
                                    ┌─────────────────┐
                                    │   ML Model      │
                                    │   (.pkl file)   │
                                    └─────────────────┘

🐳 Quick Start with Docker

Prerequisites

• Docker and Docker Compose installed
• Git

1. Clone the repository

git clone https://github.com/omarsinno54/insurance-cost-predictor.git
cd insurance-cost-predictor

2. Build and run with Docker Compose

docker-compose up --build

3. Access the application

• Frontend: http://localhost:3000
• API: http://localhost:5010
• Health Check: http://localhost:5010/health

🧪 Testing the API

Using curl commands:

Health Check:

curl http://localhost:5010/health

Expected response:

{
  "model_loaded": true,
  "status": "healthy"
}

Make a Prediction:

curl -X POST http://localhost:5010/predict \
  -H "Content-Type: application/json" \
  -d '{
    "age": 30,
    "sex": "male",
    "bmi": 25.5,
    "children": 2,
    "smoker": "no",
    "region": "northeast"
  }'

Expected response:

{
  "prediction": 5444.925698040062,
  "status": "success"
}

Using the Web Interface:

1. Go to http://localhost:3000
2. Fill out the insurance form with your details
3. Click "Predict Insurance Cost"
4. See your estimated annual insurance cost

📁 Project Structure

insurance-cost-predictor/
├── README.md
├── docker-compose.yml    # Multi-container orchestration
├── .gitignore
├── Makefile              # Makefile including MLflow
│
├── backend/ # Flask API backend
│   ├── Dockerfile
│   ├── requirements.txt
│   ├── app.py            # Flask API server
│   ├── model.pkl         # Trained ML model
│   └── Makefile          # Makefile used in Docker container
│
└── frontend/             # Web interface
    ├── Dockerfile
    ├── nginx.conf        # Nginx configuration
    └── index.html        # Interactive web interface

🔧 API Reference

GET /health

Check API health status.

Response:

{
  "status": "healthy",
  "model_loaded": true
}

POST /predict

Predict insurance cost for given parameters.

Request Body:

{
  "age": 30,
  "sex": "male",
  "bmi": 25.5,
  "children": 2,
  "smoker": "no",
  "region": "northeast"
}

Response:

{
  "prediction": 5444.925698040062,
  "status": "success"
}

Parameters:

• age: Integer (18-100)
• sex: String ("male" or "female")
• bmi: Float (15.0-50.0)
• children: Integer (0-10)
• smoker: String ("yes" or "no")
• region: String ("northeast", "northwest", "southeast", "southwest")

🐛 Troubleshooting

Common Issues

1. "Docker command not found"

# Check Docker installation
docker --version
docker-compose --version

# On macOS, if Docker Desktop is installed but command not found:
echo 'export PATH="/Applications/Docker.app/Contents/Resources/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

2. "Port already in use" If you see port conflict errors, change the ports in docker-compose.yml:

services:
  backend:
    ports:
      - "5011:5010"  # Use different external port
  frontend:
    ports:
      - "3001:80"    # Use different external port

Then access via http://localhost:3001 and http://localhost:5011

3. "Model file not found" Ensure your .pkl model file is in the backend/ directory with the correct name in app.py.

4. Changes not reflecting Always rebuild after making changes:

docker-compose down && docker-compose up --build

5. "Failed to load resource: net::ERR_NAME_NOT_RESOLVED" Check that the frontend JavaScript is using the correct URL format:

// Correct (with colon after http):
fetch('http://localhost:5010/predict', ...)

// Wrong (missing colon):
fetch('http//localhost:5010/predict', ...)

📝 What This Project Demonstrates

This project showcases:

• Machine Learning Model Deployment: Serving ML models via REST API
• Containerization: Docker and Docker Compose for consistent environments
• Full-Stack Development: Frontend interface + Backend API
• Data Processing: Log transformation for skewed target variables
• Production Considerations: Health checks, proper error handling, CORS

---

🛠️ Development

Model Training

The neural network model was trained with the following preprocessing:

• Target transformation: log10(charges) to handle right-skewed distribution
• Feature encoding: Categorical variables encoded appropriately

To retrain the model, feel free to explore the attached jupyter notebook notebook.ipynb, or write your own script!

📈 Performance Metrics

Metric	Value
R² Score	0.85
MAE	0.09

📊 Data Information

Training Data

• Dataset Size: 1338 rows
• Features: 6 input features
• Target: Insurance charges (USD)

Feature Importance

1. Smoker Status - Highest impact on premiums
2. Age - Strong positive correlation
3. BMI - Significant for high BMI values
4. Region - Regional cost variations
5. Children - Moderate impact
6. Sex - Minor statistical difference

Using Makefile

To have cleaner structure, CLI commands were set in a Makefile

# Run application
make app

# Format code
make format

# Install dependencies
make install

# Run mlflow server
make mlflow

Using MLflow

The project includes MLflow to log model hyperparameters, metrics and artifacts. The MLflow UI provides an impressive dashboard to compare model performance.

👨‍💻 Author

Omar Sinno

• GitHub: @omarsinno54
• LinkedIn: @omarhsinno
• Email: omarsinno54@gmail.com

🙏 Acknowledgments

• Dataset: Kaggle @"mosapabdelghany/medical-insurance-cost-dataset"

DOI Citation

@misc{mosap_abdelghany_2025,
	title={Medical Insurance Cost Dataset},
	url={https://www.kaggle.com/dsv/12853160},
	DOI={10.34740/KAGGLE/DSV/12853160},
	publisher={Kaggle},
	author={mosap abdelghany},
	year={2025}
}

License

CC0: Public Domain

---