Consensus feature shortlist

Build a table: feature × (domain relevance, quality, correlation/MI/IV, redundancy, leakage risk, stability) → keep/drop/transform. Concepts: Evidence-based feature review Project A: Consensus voting (your Cell 40): MI > 0.02, RF importance > 0.01, RFE top 15, IV ≥ 0.10 → keep features with ≥ 2 votes. Run all four on the TRAIN split only. Project B: Shortlist calendar, lag, rolling and weather features for the regression module.