A heart-disease risk tool that gives someone an early, data-driven warning before they see a doctor. Trained on the CDC BRFSS 2020 dataset of over 300,000 US health records and delivered through an interface a non-specialist can actually read.
What it does
- An XGBoost model over 17 health parameters, validated with K-fold cross-validation and reaching 91.5% prediction accuracy.
- A Streamlit interface that presents risk on a green-to-red scale with recommendations matched to the band, rather than a bare probability.
- Chosen for tabular health data specifically — gradient-boosted trees still outperform more fashionable approaches on this shape of data.
Worth calling out
A number is not an answer
A 63% risk score means nothing to most people. The tool translates it into a zone and a next step, which is the difference between a model and something a person can act on.
Have something that needs building?