misalisu-ai/nigeria-maize-weather-yield-model

# 🌽 Nigeria Maize Yield Intelligence > **Weather-based machine learning for state-level maize yield analysis, robustness testing, uncertainty estimation, and explainable decision support across Nigeria.** ## 📌 Project Overview **Nigeria Maize Yield Intelligence** is an end-to-end machine-learning project combining Nigerian state-level maize yield records with NASA POWER weather data. It serves two complementary purposes: 1. **3MTT Capstone:** an interactive Streamlit application for exploring yield, weather patterns, model performance, controlled climate-stress scenarios, and model explanations. 2. **Research:** a reproducible study of temporal and geographic generalization, robustness, predictive uncertainty, feature dependence, and explainability in weather-only maize-yield prediction. > **Research question:** How well can weather-only machine-learning models generalize across Nigerian states and future years, and how does predictive reliability change under controlled climate perturbations? --- ## 🚀 Live Application **Streamlit App:** `nigeriamaizeyield.streamlit…` Run locally: ```bash streamlit run app/app.py ``` --- ## 🎯 Problem Maize yield varies across Nigerian states and years. Weather conditions such as rainfall, temperature, dry spells, humidity, and solar radiation may provide useful predictive information, but models can fail when asked to predict a future year, an unseen state, or weather conditions outside their training distribution. This project therefore evaluates not only prediction accuracy, but also **generalization, robustness, uncertainty, and explainability**. --- ## 💡 Solution The dashboard allows users to: - explore state-level maize yield histories; - inspect rainfall and temperature patterns; - compare Ridge, Random Forest, and LightGBM models; - examine temporal and state-held-out performance; - explore controlled daily-weather climate-stress scenarios; - inspect predictive uncertainty; - view SHAP-base …