Insurance Policy Renewal Predictor

A FastAPI-based service that predicts whether a customer will renew their insurance policy, leveraging ensemble machine learning models trained on the “Customer Insurance Policies Prediction” dataset.

---

Table of Contents

• Project Structure
• Setup
• API Usage
• Docker
• Results

---

Project Structure

.
├── Dockerfile
├── LICENSE
├── Makefile
├── README.md
├── data
│   └── dataset.csv
├── models
│   ├── bagging_model.pkl
│   └── boosting_model.pkl
├── notebooks
│   └── pipeline.ipynb
├── requirements.txt
├── temp
└── webapp
    └── app.py

• data/: Raw CSV dataset.
• models/: Serialized joblib models (Random Forest & Gradient Boosting).
• notebooks/: EDA and model training pipeline.
• webapp/: FastAPI application source code.
• Dockerfile, Makefile: Containerization and automation scripts.

---

Setup

1. Clone the repository:

   git clone https://github.com/your-org/insurance-policy-predictor.git
   cd insurance-policy-predictor

2. Create and activate a virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the API locally:

   uvicorn webapp.app:app --reload

---

API Usage

Health Check

• Endpoint: GET /
• Response:

  { "message": "Insurance Policy Renewal Prediction API" }

Predict Renewal

• Endpoint: POST /predict

• Content-Type: application/json

• Request Example:

  {
    "age": 30,
    "driving_license": 1,
    "region_code": 10,
    "previously_insured": 0,
    "annual_premium": 25000.0,
    "policy_sales_channel": 152,
    "vintage": 100,
    "gender": "Male",
    "vehicle_age": "< 1 Year",
    "vehicle_damage": "No",
    "model_type": "bagging"
  }

• Response Example:

  {
    "model": "bagging",
    "prediction": 1,
    "probability": 0.87
  }

• Interactive API docs: Visit /docs after starting the server.

---

Docker

To build and run the application using Docker:

docker build -t insurance-predictor .
docker run -p 8000:8000 insurance-predictor

---

Results

After hyperparameter tuning:

• Random Forest (Bagging): ~90.7% accuracy
• Gradient Boosting: ~83.5% accuracy

---