ankita10yadav10/Price-Forecasting-Challenge

Machine learning solution for forecasting average weekly dry maize prices in Kenya to support smallholder farmers using agriBORA’s warehouse receipt system. # agriBORA Commodity Price Forecasting Challenge Forecasting weekly wholesale dry white maize prices for the Zindi **agriBORA Commodity Price Forecasting Challenge** using a reproducible Python pipeline and a compact EDA summary. ## What’s included - `eda_summary.md` for the competition-focused EDA write-up - `generate_submission.py` for generating the submission file - `forecast.py` for the model and feature pipeline ## Main idea The project predicts `Dry Maize__White Maize` for: - Kiambu - Kirinyaga - Mombasa - Nairobi - Uasin-Gishu The modeling workflow combines: - county-level lag features - regional/global lag features - rolling mean and volatility signals - cyclical calendar encodings - ratio and trend features ## Quick start ```bash python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt python3 generate_submission.py ``` ## Project layout ```text . ├── reports/ │ └── eda_summary.md ├── scripts/ │ └── generate_submission.py ├── src/ │ └── maize_price_prediction/ │ ├── __init__.py │ └── forecast.py ├── scrape_kamis_v2.py └── requirements.txt ``` ## Notes - The full notebook is kept in the local working copy for analysis. - Large raw datasets and generated artifacts are excluded from the public GitHub copy to keep the repo light and easy to review. - The script entry point regenerates `outputs/submission.csv` from the processed training panel when the required local data files are present. ## Competition link - agriBORA Commodity Price Forecasting Challenge