Status: In progress.

Objective

Build climate-risk features that correlate with corn futures behaviour.

Here the features are what gets scored, not the forecast. The work is mostly feature engineering, and the model is secondary.

Approach

1. F-score feature selection to shortlist the climate signals that are worth keeping.
2. Correlation-based signal weighting to combine the survivors into one risk feature.
3. Offline validation against my own mock of the competition's correlation metric, so I'm not spending submissions to find out if something works.

Where it's at

• ✅ Mock correlation metric for offline validation
• 🔄 Feature selection and signal weighting