🏥 Insurance Premium Prediction API & Serving System

An end-to-end Machine Learning application that predicts insurance premium categories based on customer demographics, lifestyle factors, health indicators, and financial information.

The project follows a complete MLOps-inspired workflow:

• Data Analysis & Feature Engineering
• Random Forest Classification Model
• FastAPI REST API Development
• Pydantic Request Validation
• Interactive Streamlit Frontend
• Cloud Deployment on Render

---

🚀 Live Demo

API Documentation

https://insurance-premium-prediction-api-serving.onrender.com

Streamlit Application

https://insurance-premium-prediction-api-serving-98v3.onrender.com/

Getting Started

1. Open the API documentation link first.
2. Wait for the API to respond successfully.
3. Open the Streamlit application.
4. Enter the required details and generate a prediction.

Note

The backend is hosted on Render and may enter a sleep state after periods of inactivity. The first request may take a little longer while the server wakes up. Once initialized, the application should respond normally.

🐳 Docker Image

The project is also available as a Docker image on Docker Hub.

Docker Hub Repository

https://hub.docker.com/r/omikalix/insurance-premium-app

Pull Image

docker pull omikalix/insurance-premium-app

Run Container

docker run -p 8000:8000 -p 8501:8501 omikalix/insurance-premium-app

Access Application

• Streamlit UI: http://127.0.0.1:8501
• FastAPI Docs: http://127.0.0.1:8000/docs

Note: The services listen on 0.0.0.0 inside the container but should be accessed through 127.0.0.1 or localhost from your browser.

---

📸 Application Screenshots

🖥️ Interactive Streamlit Dashboard

🚀 FastAPI Documentation & Testing

---

📌 Project Overview

Insurance companies determine premium categories based on multiple factors such as:

• Age
• BMI
• Smoking Habits
• Income
• Occupation
• City Tier
• Lifestyle Risk

This project leverages Machine Learning to automate premium category prediction.

The model classifies customers into:

Category	Meaning
🟢 Low	Lower insurance risk
🟡 Medium	Moderate insurance risk
🔴 High	Higher insurance risk

---

🏗️ System Architecture

User
  │
  ▼
Streamlit UI
  │
  ▼
FastAPI REST API
  │
  ▼
Random Forest Model (.pkl)
  │
  ▼
Prediction Response

---

📂 Project Structure

serving-ml-models/
│
├── images/
│   ├── low.png
│   ├── medium.png
│   ├── high.png
│   └── background_image.png
│
├── model.ipynb
├── insurance_premium_model.pkl
├── insurance_premium_api.py
├── UI.py
├── Indian_Insurance_Data.csv
│
├── requirements.txt
└── README.md

---

📊 Dataset Features

Original Features

Feature	Description
age	Customer age
weight	Weight in kg
height	Height in meters
income_lpa	Annual income
smoker	Smoking status
city	Customer city
occupation	Occupation

---

⚙️ Feature Engineering

The following custom features were created:

BMI

BMI = weight / (height ** 2)

Income Per Age

income_lpa / age

Weight Height Ratio

weight / height

Income × BMI

income_lpa * bmi

Smoker Income

smoker * income_lpa

Health Risk Index

(bmi * 0.4) + (age * 0.3) + (smoker * 20)

City Tier Classification

Tier 1
Tier 2
Tier 3

---

🤖 Machine Learning Model

Algorithm

RandomForestClassifier()

Data Preprocessing

• One Hot Encoding
• Feature Engineering
• Pipeline Architecture
• Column Transformer

Train Test Split

80% Training
20% Testing

---

📈 Model Performance

Classification Report

Class	Precision	Recall	F1
High	0.91	0.90	0.91
Low	0.91	0.93	0.92
Medium	0.88	0.87	0.87

Overall Metrics

Accuracy        : 89.63%
Macro F1 Score  : 0.90

Cross Validation

Fold 1 : 89.25%
Fold 2 : 89.63%
Fold 3 : 87.25%
Fold 4 : 90.75%
Fold 5 : 91.75%

Mean CV Accuracy : 89.73%

---

🔥 Feature Importance

Top contributing features:

1. Health Risk Index
2. BMI
3. Age
4. Weight Height Ratio
5. Smoker Income
6. Income (LPA)
7. Income × BMI
8. Income Per Age

Key Insight:

Insurance premium category is primarily influenced by:

• Health Risk
• BMI
• Age
• Smoking Status
• Income

while City Tier and Occupation contribute less to prediction.

---

⚡ FastAPI Backend

The backend API is developed using:

• FastAPI
• Pydantic
• Uvicorn

Features

• Request Validation
• JSON Response
• Interactive Swagger Documentation
• Fast Prediction Serving

API Endpoint

POST /predict

Sample Request

{
  "age": 30,
  "weight": 70,
  "height": 1.75,
  "income_lpa": 12,
  "smoker": 0,
  "city": "Pune",
  "occupation": "Software Engineer"
}

Sample Response

{
  "predicted_insurance_premium_category": "Low"
}

---

🎨 Streamlit Frontend

The project includes an interactive Streamlit application featuring:

• Modern Insurance Dashboard
• Dynamic Background Design
• Interactive Form Inputs
• Real-Time Predictions
• Personalized Recommendations
• Category-Based Visual Cards

Premium Category Visualization

🟢 Low Premium

• Low Risk
• High Savings
• Healthy Lifestyle

🟡 Medium Premium

• Balanced Risk
• Balanced Coverage

🔴 High Premium

• Higher Risk
• Enhanced Protection

---

💡 Personalized Recommendation Engine

The application generates dynamic recommendations based on:

• BMI
• Age
• Income
• Occupation
• Smoking Status
• Predicted Premium Category

Examples:

🚭 Quit smoking to reduce risk.

🏃 Maintain a healthy BMI.

🩺 Schedule annual checkups.

💰 Build an emergency healthcare fund.

---

🛠️ Installation

Clone Repository

git clone https://github.com/yourusername/insurance-premium-prediction.git

Move into Project

cd insurance-premium-prediction

Install Dependencies

pip install -r requirements.txt

---

▶️ Run FastAPI Server

uvicorn insurance_premium_api:app --reload

API Docs

http://127.0.0.1:8000/docs

---

▶️ Run Streamlit UI

streamlit run UI.py

---

☁️ Deployment

This project is fully deployed on Render.

Services

✅ FastAPI Backend

✅ Streamlit Frontend

Deployment Stack

• Render
• FastAPI
• Streamlit
• Uvicorn
• Random Forest Model

---

👨‍💻 Author

OMKAR SAWANT

Building intelligent machine learning systems with scalable deployment architectures.

Dataset: Health Insurance Dataset by Parth Patel
Source: https://www.kaggle.com/datasets/patelparth3399/indian-insurance-premium-prediction-dataset

License: MIT

Special thanks to the author for making the dataset publicly available.