An exploratory data-analysis and machine-learning project investigating how the cost of a healthy diet varies across countries, regions and time.
The analysis covers 1,379 observations, representing 175 countries between 2017 and 2024.
Access to an affordable healthy diet is an important indicator of global food security and economic wellbeing.
This project explores:
- How healthy-diet costs are distributed globally
- How costs differ across geographic regions
- How prices changed between 2017 and 2024
- Which variables are associated with higher diet costs
- How effectively regression models can estimate healthy-diet cost
The dataset contains country-level healthy-diet cost estimates expressed in purchasing-power-parity-adjusted US dollars.
Main variables include:
| Variable | Description |
|---|---|
country |
Country name |
region |
Geographic region |
year |
Observation year |
cost_healthy_diet_ppp_usd |
Daily cost of a healthy diet in PPP USD |
annual_cost_healthy_diet_usd |
Estimated annual healthy-diet cost |
cost_vegetables_ppp_usd |
Estimated vegetable component cost |
cost_fruits_ppp_usd |
Estimated fruit component cost |
cost_category |
Assigned cost category |
Dataset source: Cost of Healthy Diet by Country, 2017–2024
- Data loading and quality auditing
- Missing-value and duplicate inspection
- Exploratory data analysis
- Regional and temporal comparisons
- Feature engineering
- Regression-model training
- Model evaluation and interpretation
Initial analysis found:
- The mean daily cost of a healthy diet was approximately $3.68 PPP
- Recorded values ranged from approximately $1.70 to $8.39
- The dataset contains observations from Africa, Asia, Europe and the Americas
- Diet costs vary substantially across both countries and regions
- Several component-cost variables contain substantial missing data and therefore require careful treatment
The following regression approaches are compared:
- Linear Regression
- Ridge Regression
- Random Forest
- Gradient Boosting
- XGBoost
- LightGBM
Models are evaluated using:
- Coefficient of determination, R²
- Root mean squared error, RMSE
- Mean absolute error, MAE
- Cross-validation performance
Model results are interpreted cautiously because country-level and time-dependent data require leakage-safe validation.
.
├── assets/ # Figures used in this README
├── data/ # Dataset and data documentation
├── notebooks/ # Complete Jupyter analysis
├── README.md
└── requirements.txt