🧠 Health Insurance Premium Cost Predictor (99% Accuracy)

Welcome to the ultimate health insurance cost prediction app! 🚀
With our dual-model strategy and smart feature engineering, we've built a system that predicts premiums with R² up to 99% and less than 4% error.

---

📂 Project Overview

This project started as a simple regression task—but after diving deep into error analysis, we uncovered that age plays a huge role in prediction accuracy.

💡 Solution? We split our models:

• 👶 Young Model (Age ≤ 25) → Linear Regression
• 🧓 Rest Model (Age > 25) → XGBoost Regressor

With the addition of the Genetical Risk feature, both models saw dramatic improvements.

---

🔍 Quick Look

📁 health-insurance-premium-cost-prediction-xgboost-99-accuracy
├── app/
│   └── app.py               ← Streamlit web app
├── artifacts/               ← Visualizations (accuracy, residuals)
├── notebooks/               ← Model development Jupyter notebooks
├── prediction_helper.py     ← Model prediction utility
├── requirements.txt         ← Python dependencies
└── README.md                ← You are here!

---

🌐 Live Demo

Check it out here: Streamlit App

🚀 Try It Out

git clone https://github.com/mehulcode12/health-insurance-premium-cost-prediction-xgboost-99-accuracy.git
cd health-insurance-premium-cost-prediction-xgboost-99-accuracy
pip install -r requirements.txt

# Run the Streamlit app
cd app
streamlit run app.py

---

🧠 Modeling Strategy

1️⃣ Initial Baseline

• Used a single Linear Regression model.
• Problem: 98% R², but error rate > 60% for young users.

2️⃣ Root Cause Analysis

• Split the dataset by age (≤ 25 vs > 25).
• Found young users skewed the prediction error.

3️⃣ Dual Model Solution

• Trained separate models for each group:
  • Young: Linear Regression
  • Rest: XGBoost (with RandomizedSearchCV tuning)
• Added Genetical Risk feature = game changer 💥

---

📊 Model Evaluation

👶 Age ≤ 25 (Linear Regression + Genetical Risk)

• R² (Train/Test): ~0.99 / 0.99
• 📉 Residuals:
  
• ✅ Accuracy:
  

🧓 Age > 25 (XGBoost + Genetical Risk)

• R² (Train/Test/CV): ~0.997 / 0.997 / 0.997
• 📉 Residuals:
  
• ✅ Accuracy:
  

---

🛠 Tech Stack

• ML Models: Linear Regression, XGBoost
• Tuning: GridSearchCV, RandomizedSearchCV
• Visualization: Matplotlib, Seaborn
• Web App: Streamlit
• Backend: Python, pandas, NumPy

---

📘 Related Files

Notebook	Description
ml_premium_prediction_young_with_gr.ipynb	Model for users ≤ 25 yrs
ml_premium_prediction_rest_with_gr.ipynb	Model for users > 25 yrs
SourceOfWork.pdf	Detailed analysis + methodology

---

🤝 Contribute

Have ideas to improve it? Open a PR or raise an issue! Contributions are welcome 🙌

---

📄 License

MIT License © 2025 Mehul

---

⭐ If you liked this project, leave a star on GitHub!