3,333 rows total (3,099 train / 234 test) — one row per county per IPC analysis period. Each row includes:

County population

The period's start year and month

The county's IPC phase and Phase 3+ percentage in its previous period (lagged, not current)

The county's most recent known cereal production and cereal gap figures

Gaps in the original crop production records (about 18% of rows) were imputed using each county's own historical average, falling back to its state average where no county history existed — the published dataset has zero missing values.

Target: food_insecurity_risk — 1 if the county is in Crisis, Emergency, or Catastrophe (Phase 3+), 0 otherwise.

Split: chronological — train through mid-2025, test on the three most recent periods (Sept 2025 – Jul 2026).

This is a genuine forecasting task, not a random shuffle.

The target variable, food_insecurity_risk, represents an aggregated food insecurity classification at county level.

The target is defined as:

1 — the county is classified as being in Crisis, Emergency, or Catastrophe (IPC Phase 3 or above).

0 — the county is classified below IPC Phase 3.