Predicting farmgate milk prices in Brazil using machine learning and time series forecasting techniques.
This project focuses on predicting the farmgate milk price (farmgate_milk_price) for Brazilian milk manufacturers with a forecasting horizon of one year. The challenge combines time series analysis with regression modeling to understand the complex dynamics affecting milk pricing in the Brazilian market.
Milk price prediction is crucial for:
- Farmers 📈 Planning production and investment decisions
- Manufacturers 🏭 Supply chain optimization and cost management
- Policy Makers 🏛️ Market regulation and agricultural policy development
- Investors 💰 Risk assessment in the dairy industry
milk-prediction/
│
├── 📊 data/
│ └── tidy_data.csv # Cleaned dataset
│
├── 📓 notebooks/
│ ├── 01-EDA.ipynb # 🔍 Exploratory Data Analysis
│ ├── 02-Regression.ipynb # 📈 Statistical regression models
│ ├── 03-ML.ipynb # 🤖 Machine learning model validation
│ ├── 04-ML_prediction.ipynb # 🤖 ML Final Predictions
│
├── 📦 pyproject.toml # Project dependencies
├── 🔒 uv.lock # Dependency lock file
└── 📖 README.md # Project documentation
After extensive experimentation and evaluation, the Bayesian Ridge Regression was selected as the final model due to:
- ✅ Superior predictive performance on validation data
- ✅ Robust handling of multicollinearity among features
- ✅ Uncertainty quantification through Bayesian approach
- ✅ Regularization benefits preventing overfitting
During the model validation phase the data from 2024 was separated as the Validation Set and data from 2015 to 2023 was used as the training set with all covariates.
Below, are the results achieved with three models:
| Metric | Linear Regression | Random Forest | Bayesian Ridge |
|---|---|---|---|
| Bias | 0.0849 | -0.0170 | 0.0677 |
| MAE | 0.0979 | 0.1129 | 0.0734 |
| RMSE | 0.1390 | 0.1281 | 0.1011 |
The Bayesian Ridge model demonstrated superior performance across key metrics:
- ✅ Lowest MAE (0.0734) 📉 Best average prediction accuracy
- ✅ Lowest RMSE (0.1011) 🎯 Superior overall prediction quality
- ✅ Moderate bias (0.0677) ⚖️ Balanced systematic error
- ✅ Robust generalization 💪 Consistent performance across time periods
Below we can see the plot of the predictions on the validation set:
| Date | Farmgate Milk Price |
|---|---|
| 2025-01-01 | 2.5245 |
| 2025-02-01 | 2.5882 |
| 2025-03-01 | 2.6609 |
| 2025-04-01 | 2.7641 |
| 2025-05-01 | 2.8132 |
| 2025-06-01 | 2.9388 |
| 2025-07-01 | 3.0034 |
| 2025-08-01 | 2.9419 |
| 2025-09-01 | 2.8464 |
| 2025-10-01 | 2.7533 |
| 2025-11-01 | 2.6891 |
| 2025-12-01 | 2.6452 |
The predictions show a seasonal pattern with:
- 📈 Peak prices in mid-year (June-July) reaching ~3.00
- 📉 Lower prices at year start/end (~2.52-2.65)
- 🔄 Consistent seasonal cyclicality reflecting market dynamics
And the historical plot with the predictions below:
Made with ❤️ for the Brazilian dairy industry