---
license: mit
language: en
tags:
- nutrition
- healthcare
- elderly-care
- regression
- xgboost
- uganda
- africa
datasets:
- uganda-elderly-nutrition
- Shakiran/UgandanNutritionMealPlanning
- dongx1997/NutriBench
metrics:
- r2
- mae
- rmse
library_name: xgboost
pipeline_tag: tabular-regression
---

# XGBoost Model for Elderly Nutrition Planning in Uganda

## Model Description

This XGBoost regression model predicts daily caloric needs for elderly individuals (aged 60+) in Uganda based on nutritional content, health conditions, regional factors, and demographic information. The model is designed to support nutrition planning, meal preparation, and healthcare decision-making for elderly care in Uganda.

### Model Details

- **Model Type:** XGBoost Regressor (Gradient Boosting)
- **Task:** Tabular Regression
- **Version:** v1.0_optimized
- **Training Date:** November 3, 2025
- **Framework:** XGBoost 2.0+
- **Language:** Python
- **License:** Apache 2.0

### Developed By

- **Organization:** Graph-Enhanced LLMs for Locally-Sourced Elderly Nutrition Planning Project
- **Project Focus:** AI-driven nutrition planning for elderly populations in Uganda
- **Contact:** [shakirannannyombi@gmail.com]

---

## Intended Use

### Primary Use Cases

1. **Nutrition Planning:** Calculate appropriate caloric intake for elderly individuals based on their health profile
2. **Meal Planning:** Support caregivers and healthcare providers in designing meal plans
3. **Healthcare Decision Support:** Assist medical professionals in nutritional assessments
4. **Research:** Enable studies on nutrition needs for elderly populations in Uganda
5. **Policy Development:** Inform nutrition policies for elderly care facilities

### Intended Users

- Healthcare providers and nutritionists
- Elderly care facilities and nursing homes
- Family caregivers
- Public health researchers
- NGOs working in elderly nutrition

### Out-of-Scope Use

- ❌ Not for children or adults under 60 years
- ❌ Not for acute medical conditions requiring immediate intervention
- ❌ Not a replacement for professional medical advice
- ❌ Not validated for use outside Uganda without regional calibration

---

## Performance

### Overall Metrics

| Metric | Training Set | Test Set |
|--------|-------------|----------|
| **R² Score** | 0.9309 | **0.6710** |
| **MAE (kcal/day)** | 1.29 | **2.84** |
| **RMSE (kcal/day)** | 1.65 | **3.60** |
| **Training Time** | 25.0 seconds | - |

### Model Ranking

Compared against 5 different models (HistGradient Boosting, XGBoost, LightGBM, MLP, GNN):

- **Overall Rank:** 🥇 #1 out of 5
- **R² Rank:** 🥇 #1 (0.6710)
- **MAE Rank:** 🥇 #1 (2.84 kcal/day)
- **RMSE Rank:** 🥇 #1 (3.60 kcal/day)

### Baseline Comparison

| Metric | Baseline Model | This Model | Improvement |
|--------|---------------|------------|-------------|
| Test R² | 0.6311 | 0.6710 | **+6.3%** |
| Test MAE | 2.998 kcal/day | 2.842 kcal/day | **-5.2%** |

### Performance Characteristics

- **Strong generalization:** R² = 0.67 indicates good predictive power
- **Low prediction error:** MAE of 2.84 kcal/day is clinically acceptable
- **Moderate overfitting:** Train-test R² gap of 0.26 (manageable with regularization)
- **Consistent predictions:** RMSE close to MAE suggests few outliers

---

## Training Data

### Dataset Overview

- **Dataset Name:** Uganda Elderly Nutrition Dataset (Enriched)
- **Total Samples:** 1,000
- **Training Samples:** 700 (70%)
- **Test Samples:** 300 (30%)
- **Split Method:** Random stratified split (seed=42)

### Features (18 total)

#### Nutritional Content (12 features)
- `Energy_kcal_per_serving` - Energy content per serving
- `Protein_g_per_serving` - Protein content (grams)
- `Fat_g_per_serving` - Fat content (grams)
- `Carbohydrates_g_per_serving` - Carbohydrate content (grams)
- `Fiber_g_per_serving` - Dietary fiber (grams)
- `Calcium_mg_per_serving` - Calcium content (milligrams)
- `Iron_mg_per_serving` - Iron content (milligrams)
- `Zinc_mg_per_serving` - Zinc content (milligrams)
- `VitaminA_µg_per_serving` - Vitamin A content (micrograms)
- `VitaminC_mg_per_serving` - Vitamin C content (milligrams)
- `Potassium_mg_per_serving` - Potassium content (milligrams)
- `Magnesium_mg_per_serving` - Magnesium content (milligrams)

#### Categorical Features (4 features)
- `region_encoded` - Geographic region in Uganda (4 regions)
- `condition_encoded` - Health condition (8 conditions)
- `age_group_encoded` - Age group (3 groups: 60-70, 70-80, 80+)
- `season_encoded` - Seasonal availability

#### Other Features (2 features)
- `portion_size_g` - Portion size in grams
- `estimated_cost_ugx` - Estimated cost in Ugandan Shillings

### Geographic Coverage

**4 Regions of Uganda:**
1. Central Uganda (Buganda)
2. Western Uganda (Ankole, Tooro, Kigezi, Bunyoro)
3. Eastern Uganda (Busoga, Bugisu, Teso)
4. Northern Uganda (Acholi, Lango, Karamoja, West Nile)

### Health Conditions Covered

**8 Common Elderly Conditions:**
1. Hypertension
2. Undernutrition
3. Anemia
4. Frailty
5. Digestive issues
6. Arthritis
7. Osteoporosis
8. Diabetes

### Age Groups

- **60-70 years:** Early elderly
- **70-80 years:** Mid elderly
- **80+ years:** Advanced elderly

### Target Variable

- **Name:** Daily Caloric Needs
- **Unit:** kcal/day
- **Range:** Typically 1,400 - 2,500 kcal/day
- **Distribution:** Approximately normal

---

## Training Details

### Hyperparameters (Optimized)

```python
{
    'n_estimators': 200,
    'max_depth': 4,
    'learning_rate': 0.05,
    'min_child_weight': 5,
    'subsample': 0.8,
    'colsample_bytree': 0.8,
    'gamma': 0,
    'reg_alpha': 0,
    'reg_lambda': 1.5
}
```

### Training Configuration

- **Objective:** Regression (minimize squared error)
- **Evaluation Metric:** R² Score, MAE, RMSE
- **Validation Strategy:** 70-30 train-test split
- **Early Stopping:** Not used (200 trees)
- **Feature Scaling:** StandardScaler applied to numeric features
- **Encoding:** Label encoding for categorical features

### Training Environment

- **Hardware:** CPU-based training
- **Training Time:** 25 seconds
- **Memory Usage:** <1 GB
- **Reproducibility:** Random seed = 42

---

## How to Use

### Installation

```bash
pip install xgboost==2.0.0 pandas numpy scikit-learn
```

### Loading the Model

```python
import pickle
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler

# Load model files
with open('xgboost_nutrition_model_20251103.pkl', 'rb') as f:
    model = pickle.load(f)

with open('xgboost_scaler_20251103.pkl', 'rb') as f:
    scaler = pickle.load(f)

with open('xgboost_label_encoders_20251103.pkl', 'rb') as f:
    label_encoders = pickle.load(f)

with open('xgboost_feature_names_20251103.pkl', 'rb') as f:
    feature_names = pickle.load(f)
```

### Making Predictions

```python
# Example input data
input_data = {
    'Energy_kcal_per_serving': 350,
    'Protein_g_per_serving': 15,
    'Fat_g_per_serving': 10,
    'Carbohydrates_g_per_serving': 45,
    'Fiber_g_per_serving': 5,
    'Calcium_mg_per_serving': 200,
    'Iron_mg_per_serving': 3,
    'Zinc_mg_per_serving': 2,
    'VitaminA_µg_per_serving': 500,
    'VitaminC_mg_per_serving': 20,
    'Potassium_mg_per_serving': 400,
    'Magnesium_mg_per_serving': 50,
    'region_encoded': 0,  # Central Uganda
    'condition_encoded': 0,  # Hypertension
    'age_group_encoded': 1,  # 70-80
    'season_encoded': 0,
    'portion_size_g': 250,
    'estimated_cost_ugx': 5000
}

# Convert to DataFrame
df = pd.DataFrame([input_data])

# Ensure correct feature order
df = df[feature_names]

# Scale features (if scaler expects it)
# Note: Check if your scaler was fit on all features or just numeric ones
# df_scaled = scaler.transform(df)

# Make prediction
predicted_calories = model.predict(df)
print(f"Predicted daily caloric needs: {predicted_calories[0]:.2f} kcal/day")
```

### Using with the API

```python
import requests

url = "http://your-api-endpoint/predict"
data = {
    "data": {
        "Energy_kcal_per_serving": 350,
        "Protein_g_per_serving": 15,
        # ... other features
    }
}

response = requests.post(url, json=data)
result = response.json()
print(f"Predicted calories: {result['prediction']['caloric_needs']:.2f} kcal/day")
```

---

## Limitations and Biases

### Known Limitations

1. **Sample Size:**
   - Only 1,000 training samples may not capture all population variability
   - Recommend caution when making predictions for rare scenarios

2. **Geographic Scope:**
   - Trained specifically on Ugandan population data
   - May not generalize well to other African countries or regions

3. **Moderate Overfitting:**
   - Train-test R² gap of 0.26 indicates some overfitting
   - Predictions should be validated against clinical guidelines

4. **Feature Dependencies:**
   - Requires accurate nutritional content data
   - Missing or incorrect features will degrade performance

5. **Temporal Validity:**
   - Trained on 2025 data
   - May need retraining as dietary patterns evolve

### Potential Biases

1. **Regional Representation:**
   - May have unequal representation across regions
   - Ensure validation across all 4 regions

2. **Health Condition Bias:**
   - Some conditions may be over/under-represented
   - Validate for less common conditions

3. **Socioeconomic Factors:**
   - Cost estimates may not reflect all economic situations
   - Consider local affordability in deployment

### Uncertainty Quantification

- **Prediction Uncertainty:** ±2.84 kcal/day (MAE)
- **Confidence Intervals:** 95% CI ≈ ±5.7 kcal/day (2 × MAE)
- **Recommended Buffer:** Add 10% safety margin for meal planning

---

## Ethical Considerations

### Fairness and Equity

- Model covers all major regions of Uganda
- Includes diverse health conditions
- Considers affordability factors
- ⚠️ Ensure equal access to technology for model deployment

### Privacy

- Model trained on aggregated data (no personal identifiers)
- Predictions do not require storage of sensitive health information
- ⚠️ Implement proper data handling in deployment

### Safety

- ⚠️ **Critical:** Model outputs should be reviewed by qualified healthcare professionals
- ⚠️ Not suitable for emergency nutritional interventions
- ⚠️ Should complement, not replace, clinical judgment

### Transparency

- Open methodology and evaluation metrics
- Feature importance available for interpretation
- Model architecture and hyperparameters disclosed

---

## Model Interpretability

### Feature Importance (Top 10)

Based on XGBoost's built-in feature importance:

1. **Energy_kcal_per_serving** - Highest importance
2. **Protein_g_per_serving** - High importance
3. **Carbohydrates_g_per_serving** - High importance
4. **age_group_encoded** - Moderate importance
5. **condition_encoded** - Moderate importance
6. **portion_size_g** - Moderate importance
7. **Calcium_mg_per_serving** - Moderate importance
8. **Fat_g_per_serving** - Low-moderate importance
9. **region_encoded** - Low-moderate importance
10. **Fiber_g_per_serving** - Low importance

*Full feature importance analysis available in model artifacts*

### Explainability

- **SHAP Values:** Can be computed for individual predictions
- **Partial Dependence Plots:** Available for key features
- **Decision Rules:** XGBoost trees can be exported for inspection

---

## Comparison with Other Models

| Model | Test R² | Test MAE | Training Time | Rank |
|-------|---------|----------|---------------|------|
| **XGBoost (This Model)** | **0.6710** | **2.84** | 25.0s | 🥇 #1 |
| LightGBM | 0.6649 | 2.88 | 0.93s | 🥈 #2 |
| HistGradient Boosting | 0.5116 | 3.42 | 0.14s | 🥉 #3 |
| GNN v2 | 0.5100 | 3.42 | 5.2s | #4 |
| MLP | -0.3035 | 5.66 | 4.5s | #5 |

**Recommendation:** Use XGBoost for best accuracy; consider LightGBM for faster inference.

---

## Updates and Maintenance

### Version History

- **v1.0_optimized (2025-11-03):** Initial release
  - Trained on 1,000 samples
  - Hyperparameter optimization completed
  - Test R² = 0.6710

### Planned Improvements

1. **Data Collection:**
   - Expand dataset to 5,000+ samples
   - Include more seasonal variations
   - Add rural vs. urban distinctions

2. **Feature Engineering:**
   - Add BMI calculations
   - Include activity level metrics
   - Incorporate cultural food preferences

3. **Model Enhancements:**
   - Ensemble with LightGBM for improved accuracy
   - Implement SHAP-based explainability
   - Add prediction uncertainty intervals

4. **Validation:**
   - Clinical validation studies
   - Cross-regional performance assessment
   - Temporal validation (seasonal changes)

### Retraining Schedule

- **Recommended:** Every 6-12 months
- **Triggers:** New data availability, significant dietary changes, performance degradation

---

## Citation

If you use this model in your research or application, please cite:

```bibtex
@misc{uganda_elderly_nutrition_xgboost_2025,
  title={XGBoost Model for Elderly Nutrition Planning in Uganda},
  author={[Your Name/Organization]},
  year={2025},
  month={November},
  howpublished={Hugging Face Model Hub},
  url={https://huggingface.co/[your-username]/xgboost-elderly-nutrition-uganda}
}
```

---

## Additional Resources

### Related Links

- **Project Repository:** [https://github.com/Shakiran-Nannyombi/Graph-Enhanced-LLMs-for-Locally-Sourced-Elderly-Nutrition-Planning-in-Uganda.git]
- **API Documentation:** [API Docs Link]
- **Research Paper:** [Paper Link if available]
- **Dataset:** [Shakiran/UgandanNutritionMealPlanning]

### Model Artifacts

- `xgboost_nutrition_model_20251103.pkl` - Main XGBoost model
- `xgboost_scaler_20251103.pkl` - Feature scaler (StandardScaler)
- `xgboost_label_encoders_20251103.pkl` - Categorical encoders
- `xgboost_feature_names_20251103.pkl` - Feature name list
- `xgboost_model_metadata_20251103.json` - Complete metadata

### Support

For questions, issues, or contributions:
- **Issues:** [https://github.com/Shakiran-Nannyombi/Graph-Enhanced-LLMs-for-Locally-Sourced-Elderly-Nutrition-Planning-in-Uganda.git]
- **Email:** [devkiran256@gmail.com]
- 
---

## License

This model is released under the **Apache License 2.0**.

- Commercial use allowed
- Modification allowed
- Distribution allowed
- Patent use allowed
- ⚠️ Must include license and copyright notice
- ⚠️ Must state significant changes

**Disclaimer:** This model is provided "as is" without warranty. Users are responsible for validating the model's suitability for their specific use case and ensuring compliance with local healthcare regulations.

---

## Acknowledgments

### Data Sources and References

This model was developed using knowledge and data extracted from the following authoritative sources:

1. **Handbook_Eldernutr_FINAL.pdf**
   - Comprehensive handbook on elderly nutrition
   - Primary reference for nutritional requirements and guidelines

2. **WHO ICOPE Guidelines (icope.pdf)**
   - World Health Organization Integrated Care for Older People (ICOPE)
   - Framework for elderly healthcare and nutrition assessment

3. **Nutritional_Requirements_of_Older_People.pdf**
   - Detailed nutritional requirements for elderly populations
   - Evidence-based dietary recommendations

4. **TipSheet_21_HealthyEatingForOlderAdults.pdf**
   - Practical tips for healthy eating in older adults
   - Community-oriented nutrition guidance

5. **MSD Manual Professional Edition**
   - "Drug Categories of Concern in Older Adults - Geriatrics"
   - Clinical reference for medication-nutrition interactions

6. **MSD Manual Consumer Version**
   - "Aging and Medications - Older People's Health Issues"
   - Patient-friendly information on aging and health

7. **Uganda Nutrition Data (download.pdf)**
   - Uganda-specific nutritional data and food composition
   - Local context and dietary patterns

8. **Street Food Nutritional Analysis**
   - "Average energy and nutrient contents of typical street food dishes in Uganda (Kampala)"
   - Local food nutritional profiles for urban Uganda

### Institutional Support

- **Uganda Ministry of Health** - Nutrition guidelines and policy frameworks
- **World Health Organization (WHO)** - ICOPE framework and elderly care guidelines
- **MSD Manuals** - Clinical and consumer health information

### Technical Contributions

- **Open-source community:** XGBoost, scikit-learn, pandas, Python ecosystem
- **Healthcare professionals** who contributed domain expertise
- **Data scientists and researchers** in elderly nutrition and machine learning

### Regional Knowledge

- Local nutrition experts from Uganda's 4 major regions:
  - Central Uganda (Buganda)
  - Western Uganda (Ankole, Tooro, Kigezi, Bunyoro)
  - Eastern Uganda (Busoga, Bugisu, Teso)
  - Northern Uganda (Acholi, Lango, Karamoja, West Nile)

### Special Thanks

- Community health workers providing ground-level insights
- Elderly care facilities participating in data validation
- Nutrition researchers focusing on African elderly populations
- Open data initiatives promoting nutrition research in Uganda

---

**Last Updated:** November 4, 2025
**Model Version:** v1.0_optimized
**Status:** Production Ready