What might your medical insurance cost?

An estimate with a range, the changes that would move it most, and accuracy numbers measured without the data leak that inflated the original project.

About you

Do you smoke?
Sex & region (not used for pricing)

The model ignores these — removing them cost no accuracy (see How accurate?). Kept only so the form mirrors the dataset.

Estimated annual charges

—
Fill in the form to get an estimate.

What would change it?

    How accurate is this? the gap we fixed

    Same untouched test set, three ways of scoring

    All models tried (5-fold cross-validation, repeated 3×, on unique rows only)

    About

    MedCost is a rebuilt version of Medical-Insurance-Cost-Predictor. What changed: duplicates removed before splitting; honest cross-validated model selection; a smoker × obesity feature; a gradient-boosting model with a robust (Huber) loss; 90% prediction ranges calibrated separately for smokers and non-smokers; what-if scenarios; warnings outside the training range; sex and region dropped from pricing.

    Educational tool trained on a small public US dataset (1,337 people, ages 18–64). Not a quote, not financial or medical advice.