Trizziee/African-Weather-Impact-Predictor-on-Crop-Yield-

# 🌾 Africa-Wide Crop Yield & Climate Analysis **APT3010A — Introduction to Artificial Intelligence | Group** Project 9: *Weather Impact Predictor on Agricultural Yield* Predicts crop yield (kg/ha) across African countries using 25 years (2000–2024) of FAOSTAT agricultural records combined with NASA POWER monthly climate data, and ships as a working prediction tool an extension officer or farmer can actually click through — no notebook required. ## Deployed site: uxtpode3fo94gwpsbazskm.stre… ## What's in this repo | Path | What it is | |---|---| | `G7_Africa_Crop_Yield_Climate_Analysis.ipynb` | Full analysis: data collection, cleaning, feature engineering, model training/comparison, evaluation, and model export | | `crop_yield_app/app.py` | Streamlit UI wrapping the trained model in a clickable prediction tool | | `crop_yield_app/requirements.txt` | Python dependencies for the app | | `*.joblib` | Trained model + supporting lookup tables, exported by the notebook (Step 13) | ## Project overview - **Data sources:** FAOSTAT crop yield records (Maize, Wheat, Rice, Beans, Sorghum, Potatoes, Sugar cane) and NASA POWER monthly temperature/rainfall, pulled for 50 African countries using capital-city coordinates as a country-level proxy. - **Unit of analysis:** Country × Crop × Year. - **Feature engineering:** seasonal indicators derived from monthly data (peak-quarter rainfall, rainfall variability, temperature range) plus country-crop-specific anomalies and prior-year lag features. - **Modeling:** Linear Regression, Decision Tree, Random Forest, and Gradient Boosting compared on a **country-grouped** train/test split (evaluated on countries the model never saw during training, not a random split). - **Deliverable:** a prediction tool where a user picks a country and crop, enters expected seasonal temperature and rainfall, and gets a predicted yield, a comparison to that country-crop's historical average, and a plain-language outlook (drought / hot / …