Logo Lanfrica

marcoBorto2921/agriBORA-Commodity-Price-Forecasting-Challenge

Domaine:

agriculturesocioeconomic

Type de record:

model
Créateur:
mar
Hôte:
Unofficial #1 solution for the agriBORA Zindi competition — weekly maize price forecasting (KES/kg) for 5 Kenya counties. XGBoost + county-specific carryforward blend. Private LB +0.820 vs official winner +0.801. # agriBORA Commodity Price Forecasting Challenge > **Unofficial Rank #1** — this solution was developed as a retrospective exercise after the competition closed (December 27, 2025), with full access to the ground truth. It is not an official competition result. > > The winning score on the official private leaderboard was **+0.801**. This solution scores **+0.820** on the same private test set — which would have ranked #1 had it been submitted during the competition. > > In Zindi competitions the private leaderboard score is the definitive result: it is computed on the full hidden test set, while the public leaderboard covers only a subset and is prone to overfitting. Retrospective solution for the agriBORA Commodity Price Forecasting Challenge on Zindi (closed December 27, 2025). --- ## Problem Forecast weekly wholesale maize price (KES/kg) for 5 Kenya counties — Kiambu, Kirinyaga, Mombasa, Nairobi, Uasin-Gishu — for Week 52 (Dec 22, 2025) and Week 1 (Dec 29, 2025). **Metric:** `0.5 × MAE + 0.5 × RMSE` — lower is better. Zindi normalizes to higher=better on the leaderboard. **Submission:** 10 rows (5 counties × 2 weeks). Columns `Target_RMSE` and `Target_MAE` must be identical. --- ## Dataset | File | Description | |------|-------------| | `agriBORA_maize_prices.csv` | agriBORA training series (Oct 2023 – Oct 2025) | | `agriBORA_maize_prices_weeks_46_to_51.csv` | Rolling ground truth Wk 46–51 (released during competition) | | `agriBORA_Final_Weeks_maize_price.csv` | Ground truth Wk 52 + Wk 1 (sealed until end) | | `kamis_maize_prices.csv` | KAMIS supplemental data: retail/wholesale prices + supply volumes, May 2021 – Jul 2025 | Data is not redistributable. Download from the Zindi competition page. **Key data characteristics:** - Uasin-Gishu: ~60 obs — best covered, major producer county - Nairobi: ~45 obs — major consumer hub, strong AR(1) structure (β=0.861) - Kiambu: ~25 obs — peri-urban, augmented with KAMIS - Kirinyaga: ~20 obs — Mt. Kenya region - …