Adedeji-Taiwo/Cocoa-Quality

An interactive ML-powered decision-support tool for cocoa farmers and processors in West Africa. Predicts fermentation index from field conditions and recommends optimal fermentation duration. # CocoaQuality – Fermentation Quality Predictor > An interactive ML-powered decision-support tool for cocoa farmers and processors in West Africa. Predicts fermentation index from field conditions and recommends optimal fermentation duration. **Developed as part of the Advanced Analytics for Agribusiness course, MSc in Agribusiness & Innovation, UM6P.** --- ## Live Demo Launch App on Streamlit Cloud --- ## Problem Statement Cocoa fermentation is the most critical post-harvest step determining bean quality and farmer income. Poor fermentation — either under or over — directly reduces the market grade and price received. Farmers in Nigeria, Ghana, and Côte d'Ivoire often rely on experience alone, with no quantitative tool to guide decisions. This app provides a data-driven recommendation engine that tells farmers: - What quality grade their current batch will achieve - Exactly when to stop fermentation for maximum value - How much revenue they are leaving on the table vs the optimal scenario --- ## Features | Feature | Description | |---|---| | Live weather fetch | Auto-fills temperature and humidity from any West African city via Open-Meteo API (no key required) | | Animated quality curve | Shows predicted fermentation index across days 2–8 with uncertainty band | | Optimal day marker | Identifies the peak quality day for current conditions | | Revenue calculator | Converts quality grade to CFA/kg price and computes batch value vs optimal | | West Africa heatmap | Choropleth map of fermentation quality potential by country | | Model comparison | RMSE / MAE / R² for all 3 models with feature importance chart | --- ## ML Pipeline ### Models compared | Model | RMSE | R² | |---|---|---| | Linear Regression | ~0.14 | ~0.92 | | Random Forest | ~0.12 | ~0.95 | | Gradient Boosting ★ | ~0.11 | ~0.95 | Gradient Boosting selected as best model. Random Forest retained for uncertainty quantification (variance across trees). ### Features - Fermentation duration (d …