Mrisaac7979/togo-crop-recommendation-ml

Prototype ML - Crop recommendation system for Togolese smallholder farmers # Crop Recommendation System for Togolese Smallholder Farmers (Methodological Prototype) Machine learning prototype developed in preparation for the Master's research project **"Intelligent Decision Support for Togolese Smallholder Farmers: A Machine Learning Approach to Agricultural Decision Support in Low-Resource Contexts"** **Author:** Komi Isaac Junior Hounbo ## Objective Build a complete machine learning pipeline to recommend the most suitable crop (**Maize, Cassava, or Soybean**) for a given plot, based on agronomic variables: soil properties (pH, nitrogen, phosphorus, clay content), climate variables (cumulative rainfall, average temperature), and water stress index. This use case corresponds to **Use Case 1** (Phase 2 - Modeling) of the research project: crop variety recommendation via Random Forest / XGBoost on tabular data combining soil profiles, rainfall forecasts, and historical yield data. ## Data The dataset used (`dataset_recommandation_culture.csv`) is **synthetic**, generated to reproduce the realistic statistical structure of agronomic variables typical of the Maritime and Plateaux regions of Togo. It serves to **validate the methodological pipeline** before applying it to the real data to be collected in Phase 1 of the Master's project: - Surveys of 150–200 farmers (in collaboration with ITRA) - Soil profile data (ITRA agronomic database) - Satellite imagery from Sentinel-2 / MODIS (Google Earth Engine) — NDVI indices - Ten-year climate time series (Togo's National Meteorological Directorate) ## Methodology 1. Exploratory data analysis (variable distributions per crop, correlation matrix) 2. Feature engineering: composite soil fertility index (nitrogen, phosphorus, pH) 3. Target encoding and stratified train/test split (80/20) 4. Modeling: Random Forest with hyperparameter tuning (GridSearchCV, 5-fold cross-validation) 5. Evaluation: accuracy, weighted F1-score, confusion matrix 6. Interpretation: feature importance analysis ## Results …