mah-trigui/agriBORA-Commodity-Price-Forecasting-Zindi-Competiton

Forecast weekly market prices for maize in Kenya # Kenya Maize Price Forecasting — Pipeline This competition is hosted on Zindi, a machine learning platform for data science challenges. Here is the link to the competition: agriBORA Commodity Price Forecasting Challenge 🌾 - Win €8,250 EUR Ranked in the TOP 24% --- Zindi competition · agriBORA × KAMIS · Weekly price prediction for 5 Kenya counties. --- ## Competition Overview | Item | Details | |---|---| | Task | Predict the weekly average wholesale maize price (KES) | | Target counties | Kiambu, Kirinyaga, Mombasa, Nairobi, Uasin-Gishu | | Horizon | 2 weeks ahead, rolling | | Metric | `score = 0.5 × MAE + 0.5 × RMSE` (lower is better) | | Submission format | One row per `{County}_Week_{iso_week}` | | Both target columns | `Target_RMSE` and `Target_MAE` hold the same predicted price | | Seed | `1618` | --- ## Data Sources | File | Description | |---|---| | `kamis_maize_prices.csv` | KAMIS daily market prices — filtered to `Classification == "White_Maize"` | | `agribora_maize_prices.csv` | agriBORA weekly transaction prices (main series) | | `agriBORA_maize_prices_weeks_46_to_51.csv` | Extended agriBORA weeks 46–51 | | `agriBORA_Final_Weeks_maize_price.csv` | Final scoring truth (optional) | | `SampleSubmission.csv` | Submission format reference | All data files must be placed in `DATA_DIR` (default: `C:/Users/mtrigui2/Desktop/Z1`). --- ## Pipeline Structure ``` pipeline_maize/ ├── 00_config.r Libraries, paths, constants, helper functions ├── 01_data_loading.r Load KAMIS + agriBORA raw data ├── 02_data_cleaning.r Snap to Monday, impute, calibrate, LOCF fill ├── 03_feature_engineering.r Lags, rolling stats, seasonality, momentum ├── 04_build_train_test.r Multicollinearity removal, chronological split ├── 05_models.r LightGBM, XGBoost, ARIMA, ETS, Prophet, Ensemble ├── 06_evaluation.r Walk-forward CV, holdout leaderboard, feature importance ├── 07_submission.r Rolling 2-wee …