Health Insurance Premium Predictor

Live Demo & API (If the URLs stop working, please restart them)

• 🚀 Live Testing Link: Health Insurance Premium Predictor
• 🖥 Backend API: Insurance Predictor API
• 🎮 Play with Data: Interactive Data Exploration

This project integrates the entire machine learning pipeline for predicting health insurance premiums using multiple models, extensive data analysis, and interactive UI.

---

📌 Project Journey & Contribution Guidelines

If you wish to contribute, follow this step-by-step guide.

---

🔍 1. Data Analysis & Preprocessing

• Conducted extensive exploratory data analysis (EDA) using scatter plots, box plots, 3D visualizations, KDE plots, heatmaps, and more.
• Interactive data visualizations available here.
• Feature Engineering:
  • One-hot encoding for categorical variables (sex, smoker, region).
  • Log transformation applied to target variable charges using np.log1p().
• Dataset Cleaning: Handled missing values, outliers, and performed feature scaling.
• Libraries Used: pandas, numpy, seaborn, matplotlib, scikit-learn, optuna, streamlit, gradio, huggingface
• Detailed analysis available in insurance_data_analysis.ipynb.

---

🤖 2. Model Selection & Performance Evaluation

A. Preprocessing & Model Testing

Tested multiple models: XGBoost, Decision Tree, Random Forest, Linear Regression, Polynomial Regression.

B. Linear Regression

Trained after preprocessing, yielding:

• Mean Squared Error (MSE): 0.1746
• Scaled MSE (MSE/2): 0.0873
• Mean Absolute Error (MAE): 0.2685

C. Polynomial Regression (Degree = 2)

After testing multiple degrees, best results with degree = 2:

• MSE: 0.1207
• Scaled MSE: 0.0604
• MAE: 0.2049

Saved in insurance_parametric_regression.ipynb.

D. Decision Tree & Random Forest

Hyperparameter tuning applied using Optuna.

Decision Tree Results:

• MSE: 0.1464
• Scaled MSE: 0.0732
• MAE: 0.2188

Random Forest Results:

• MSE: 0.1302
• Scaled MSE: 0.0651
• MAE: 0.2073

(See insurance_Decision_tree.ipynb for details.)

E. Final Model: XGBoost (Optimized with Optuna)

• MSE: 0.1278
• Scaled MSE: 0.0639
• MAE: 0.2022

Final model saved in XGBoost_insurance_model.ipynb.

---

⚙️ 3. MLOps & Deployment

A. Backend API (FastAPI & Gradio)

• Handles incoming user input, loads trained models, and returns predictions.
• Uses FastAPI and Gradio for deployment.
• Backend triggers on input, loads models via joblib, and returns results.
• Deployed on Hugging Face Spaces.

Backend Features:

✅ FastAPI Setup: Handles API requests and CORS. ✅ Model Loading: Supports XGBoost, Decision Tree, Random Forest, Linear Regression, and Polynomial Regression. ✅ Data Preprocessing: One-hot encoding of categorical variables. ✅ Prediction Handling: Accepts input, preprocesses, selects the model, and returns predictions. ✅ Endpoints:

• / - Root endpoint (Welcome message).
• /predict - Accepts data and returns predictions.
• /health - Health check endpoint. ✅ Logging & Error Handling: Ensures smooth debugging. ✅ Cross-Origin Compatibility: Allows frontend to communicate via CORS. ✅ Deployment: Hosted on Hugging Face Spaces.

Backend file: prediction_handler.py

---

💻 4. Frontend (Streamlit UI & Gradio)

Developed an interactive chatbot-style UI for Health Insurance Premium Prediction.

A. Technologies Used

• Streamlit - Web framework.
• Gradio - Interactive UI framework.
• HTML, CSS, JavaScript - UI enhancements.
• Fetch API - Calls backend FastAPI.
• ML Models: Supports XGBoost, Decision Tree, Random Forest, Linear Regression, and Polynomial Regression.

B. Features

🎨 Modern UI: Clean and intuitive design with custom styling. 📊 Dropdown for Model Selection: Users dynamically select models. ✅ Validation Checks:

• Ensures valid age (18-100), BMI range, and categorical values. ⏳ Loading Animation & User Interaction: Enhances user experience.

Frontend file: front_end.py

---

🔮 5. Future Improvements

🔹 Hyperparameter tuning with Bayesian Optimization 🔹 Integration with cloud platforms (AWS/GCP) for scalable deployment 🔹 Mobile-friendly UI enhancements 🔹 Adding more ML models and explainable AI (SHAP, LIME)

📢 Want to contribute? Follow the structure in the respective .ipynb files and reach out via the repository issues section!

---

📝 Final Note:
For a better understanding of the project workflow, open and explore each file. I have added detailed comments in every script and notebook to guide you through each step, from data preprocessing to model deployment. These comments will help you follow the logic behind every decision and modification. 🚀

Let me know if you'd like any further refinements!

📜 License & Acknowledgments

This project is open-source and licensed under MIT License.

Special thanks to:

• Hugging Face Spaces for hosting.
• Optuna for hyperparameter tuning.
• Streamlit & FastAPI for interactive UI and API.
• Scikit-learn, Pandas, Numpy, Matplotlib, Seaborn for data processing and visualization.
• Community Contributors for improvements.

💡 Stay tuned for updates! 🚀