FoodDesert IQ

ML app that predicts if your county is a food desert. Real USDA + CDC data, 66K census tracts, instant risk score. Making food inequality visible.

Comment

Inspiration: 19 million Americans live in food deserts but most people have no idea if their own community is one of them. I wanted to build something that made this invisible crisis visible to anyone — not just researchers or policymakers.

What I Built: FoodDesert IQ is a machine learning web app where you type in any US county and instantly get:

  • A Food Access Risk Score (0–100)
  • A Community Snapshot — poverty rate, vehicle access, SNAP recipients
  • An ML Probability Breakdown across Low, Moderate, and High risk
  • Actionable resources — food banks, SNAP eligibility, policy links

How I Built It: I merged two real government datasets:

  • USDA Food Access Research Atlas — 72,531 census tracts
  • CDC PLACES Health Data — 229,298 county health records

After cleaning and normalizing both datasets (fixing state name mismatches, normalizing county names, filling missing values), I merged them into 66,274 rows and trained a 'Random Forest Classifier' to predict obesity risk levels from food access features.

The backend is 'Flask', the frontend is vanilla HTML/CSS/JS, and the whole thing is deployed on Render.

What I Learned

  • How to clean and merge real-world government datasets
  • How to engineer features to improve ML signal
  • How to build and deploy a full-stack ML web app
  • That poverty rate and vehicle access are the strongest predictors of food insecurity risk

Challenges The biggest challenge was the state name mismatch between datasets — USDA used full names ("Illinois") while CDC used abbreviations ("IL"). This caused the initial merge to return only 192 rows instead of 66,000+. Fixing this with a state mapping dictionary was a breakthrough moment.

Built With

Share this project:

Updates