RiskyEats Intelligence

Intelligence › At-risk

Model forecast

Aggregate 90-day closure pressure

Florida closure-pressure model

481 restaurants flagged as high-risk for closure in the next 90 days, based on inspection trajectory + market signals (predictions are 15 days old) This is not a statement that these restaurants will close. It means their public-record feature trajectory matches the pattern of restaurants that historically closed within 90 days. Some will recover; some may already be on the way down.

Datestamp 2026-11-14 18:40 UTC

Named forecasts are withheld. This page currently publishes aggregate model context only. RiskyEats will not publish or index named-business closure forecasts until canonical evidence links, record-completeness rules, a model card, editorial thresholds, false-positive policy, and a correction workflow are all in place. A forecast is not an official closure determination.
Publishable high-risk rows481
Holdout AUC0.705
Calibration error0.017
Top-decile precision28.7%

Methodology

Model: LogisticRegression with class_weight='balanced', wrapped in CalibratedClassifierCV sigmoid calibration. Training rows are 90-day historical snapshots from 2014-2024 with a stratified calibrated holdout; 2025 is reserved for drift review. Features are inspection trajectory, license timing, ownership churn, Sunbiz status, cuisine-county closure pressure, high-turnover-area membership, and local saturation. Source data is DBPR public records plus RiskyEats enrichment files. Known limitation: historical Sunbiz status is not available for every snapshot, so the model treats missing historical status conservatively.

Holdout false-positive rate at 50% probability: 0.0%. Probabilities are calibrated estimates, not raw model scores; forecast intervals are approximate and widen when the data is sparse.