Analytics-for-Rural-Transformation-ART/Tanzania-Agricultural-Price-Predictions

Predicting agricultural food prices for Tanzania. # Tanzania Agricultural Price Predictions ## 📌 Overview This repository contains the analytical pipeline for modeling and predicting rural wholesale prices for 8 Crop commodities (maize, rice, bean, wheat, potato, bulrush millet, finger millet, and sorghum) across **43 markets in Tanzania**. **🔗 Rmarkdown Report:** View the full analysis and 2025 predictions here --- ## 📂 Repository Structure ### 📁 Codes/ * **`Tanzania_price_predictions-notebook.Rmd`**: This is the primary workflow. Documents model fitting, evaluation, and comparison of predicted prices against regional aggregates. * **`Data_Extraction_from_PDF_To_Excel_and_cleaning.R`**: Automates data extraction from Tanzania Ministry of Industry and Trade PDF reports. Includes cleaning of month-by-month data. * **`LNMO-Spatial-CV.R`**: Contains the **Leave-N-Markets-Out (LNMO)** cross-validation strategy, which evaluates the sensitivity of predictive accuracy as increasingly more markets are held out from training. ### 📁 Data/ * **`Tanzania_Price_Data_AllCrops_with_Coordinates_5.csv`**: This is the master dataset. It contains cleaned, compiled, and geocoded prices for all crops from **May 2021 to January 2026**. ### 📁 Outputs/ * **Spatio-temporal Maps**: These are final model outputs for all 8 crop commodities, generated through Random Forest modeling to show real price variation across space and time. --- ## 🛠️ Methodology & Scope * **Model:** Random Forest Regression and Autoregressive Random Forest. * **Validation:** Temporal split, Random split, and LNMO Cross-Validation. * **Impact:** Our framework combines periodic market reports with open spatial data (travel times, climate, population density) to provide low-cost, forward-looking price estimates across space and time. These outputs can guide targeted support measures and feed simple mobile alert systems that require price information. ---