01The problem
A trained estimator is only one part of a useful prediction system. People also need structured inputs, saved predictions and a controlled way to update the model.
02The approach
- Train and save a Random Forest Regressor with crop encoding and dataset preparation.
- Expose prediction endpoints through Flask and persist user and prediction records with SQLAlchemy.
- Provide farmer prediction history alongside administrator dataset-upload, retraining and model-version workflows.
03The evidence
- Training and prediction code are separated into ml/train_model.py and ml/predict.py.
- The application includes database migrations, farmer and administrator flows, and unit/integration tests.
04Scope & status
An application prototype. No verified field-accuracy, agronomic validation or production farming-advice claim is made here.
Public source is available for independent inspection.