02 / Applied machine learning
Insurance Charges Predictor
A trained regression model exposed through a real-time FastAPI application.
Overview
This project translates a classic regression problem into an accessible application that can return an estimate from insurance-related input attributes. It includes the data workflow, trained model, API and deployment.
How it works
I built preprocessing and feature-engineering steps, compared approaches, and trained a Gradient Boosting model in scikit-learn. I evaluated the model using R², RMSE and MAE, then exposed predictions through FastAPI endpoints and deployed the application to Render.
The deployment demonstrates the whole path from a notebook-based machine learning experiment to a system others can interact with.
What mattered
For deployed ML, serving and explaining the model can be as important as training it. The evaluation metrics describe performance on the project dataset and are not a guarantee for unseen populations.