🌾 Crop Yield Prediction with Machine Learning

This project builds a machine learning pipeline to predict crop yield (kg/ha) based on weather patterns, pesticide usage, and crop/country data. The goal is to better understand which factors influence agricultural productivity and to eventually assist farmers or policymakers with actionable insights.

---

📊 Data Sources

• FAOSTAT (Food and Agriculture Organization)
  https://www.fao.org/faostat
• World Bank Climate Data
  https://climateknowledgeportal.worldbank.org/
• Pesticide Use Dataset
  Extracted via FAOSTAT → Inputs → Pesticides

---

🧪 Models Used

We trained and compared the following regression models:

• Decision Tree Regressor
• Random Forest Regressor
• Gradient Boosting Regressor
• Support Vector Regressor (SVR)

Evaluation metrics:

• R² (Coefficient of Determination)
• RMSE (Root Mean Squared Error)
• MAE (Mean Absolute Error)

---

📁 Project Structure

ml-crop-yield-analysis/
│
├── data/
│ ├── raw/ # Raw input datasets (e.g. climate, pesticides)
│ └── processed/ # Cleaned and merged dataset
│
├── notebooks/ # Jupyter Notebooks for EDA and modeling
├── requirements.txt # Python dependencies
├── README.md # Project overview and setup instructions
└── .venv/ # (optional) Local virtual environment

📈 Key Results

After testing several models, the Random Forest Regressor gave the best results when evaluated crop-by-crop:

Crop	R²	MAE	RMSE
Barley	0.9073	363.04	541.56
Maize (corn)	0.8151	685.17	1069.66
Potatoes	0.8820	2414.69	3515.61
Rice	0.8761	464.94	740.22
Soya beans	0.8080	247.71	356.41
Wheat	0.9302	376.93	539.53

Other models (like Decision Tree and Gradient Boosting) looked promising on overall metrics but performed poorly on individual crops.

---

📌 Highlights

• End-to-end ML workflow: data cleaning, feature scaling, modeling, evaluation
• Used One-Hot Encoding for country and crop variables
• Visualized correlations, distributions, and model results
• Plotted feature importances: crop type and pesticides ranked highest
• Explored per-crop prediction accuracy using grouped metrics

---

🚀 Future Work

• Train dedicated models for each crop to improve precision
• Add features like soil type, NDVI index, or fertilizer usage
• Apply time-series modeling for yearly prediction trends
• Build a simple web app to recommend yield estimates

---

👨‍💻 Author

Glenn Noronha
📍 GitHub