Skip to content

Repository files navigation

🍽️ Child Malnutrition Risk Prediction in Chad

Python Data Models Best AUC Status

Over half of Chadian children under five are malnourished. This project builds a machine learning pipeline to predict which children are at highest risk — using 9,826 children from the DHS Chad 2014 survey — enabling NGOs and health workers to intervene before malnutrition becomes acute or fatal.


🌐 Live Demo

👉 Try the app here

Enter a child's age, weight, and height to get an instant malnutrition risk assessment — designed for community health workers and NGO field teams in Chad.


🌍 Why This Matters

Chad has one of the highest child malnutrition rates in the world:

  • 42.9% of children under five are stunted
  • 32.5% are underweight
  • 14.2% are wasted — above the WHO emergency threshold
  • 52.9% suffer from some form of malnutrition

Yet most humanitarian responses are reactive — children are identified only when they arrive at a health facility in crisis. This project asks: can we predict which children are at highest risk before they become acutely malnourished?


📊 Dataset

  • Source: DHS Program — Chad Standard DHS 2014
  • File: Children's Recode (KR) — 18,623 total children
  • Measured: 9,826 children with anthropometric measurements
  • Target: Any malnutrition (stunting OR underweight OR wasting)

Malnutrition Definitions (WHO Standards)

Indicator Column Threshold Prevalence
Stunting HW70 (HAZ) Z-score < -2 42.9%
Underweight HW71 (WAZ) Z-score < -2 32.5%
Wasting HW72 (WHZ) Z-score < -2 14.2%
Any malnutrition Combined Any above 52.9%

🔬 Methodology

DHS Chad 2014 (18,623 children)
        ↓
Filter to measured children (9,826)
        ↓
Extract malnutrition Z-scores (HW70, HW71, HW72)
        ↓
Feature selection — 28 variables across 5 domains
        ↓
Remove redundant features (correlation analysis)
        ↓
Median imputation for missing values
        ↓
7 ML classifiers benchmarked
        ↓
5-Fold Stratified Cross-Validation
        ↓
Feature importance analysis

🛠️ Features Used (24 Variables)

Domain Features
Child anthropometrics Age in months, Weight, Height
Mother health BMI, Height, Education, Age
Birth history Birth order, Preceding birth interval, Births in 5 years
Household Wealth index, Size, Water source, Toilet, Electricity
Feeding practices Breastfeeding duration, Weaning age

🤖 Model Results

Model Accuracy ROC-AUC
Gradient Boosting 92.0% 0.979
XGBoost 91.8% 0.979
CatBoost 91.3% 0.974
Decision Tree 85.1% 0.919
Logistic Regression 80.4% 0.886
Random Forest 79.2% 0.889
KNN 67.2% 0.732

All models evaluated using 5-Fold Stratified Cross-Validation on 9,826 children.


🔑 Key Findings

1. Over half of Chadian children are malnourished

52.9% of children under five in Chad suffer from stunting, underweight, or wasting — making malnutrition the default condition, not the exception.

2. Three measurements predict 95% of outcomes

Child age, weight, and height alone account for ~95% of the model's predictive power. A community health worker with a scale and measuring tape can screen children effectively.

3. Malnutrition compounds with age

Child age is the strongest predictor (importance: 0.421). Stunting accumulates over time — early intervention before age 2 is critical to prevent irreversible damage.

4. Intergenerational malnutrition is real

Mother's height predicts child stunting — malnourished mothers raise malnourished children. Breaking this cycle requires maternal nutrition programs, not just child feeding.

5. Wealth and sanitation matter less than expected

Water source, toilet type, and electricity show near-zero predictive importance — suggesting that in Chad's context, maternal and child anthropometrics are far stronger signals than household infrastructure.


📌 Feature Importance (Top 10)

Rank Feature Description Importance
1 HW1 Child age in months 0.421
2 HW2 Child weight kg 0.301
3 HW3 Child height cm 0.229
4 V438 Mother height cm 0.007
5 V445 Mother BMI 0.006
6 V191 Wealth index score 0.005
7 V221 Marriage to first birth interval 0.004
8 V133 Mother education years 0.004
9 M7 Duration of breastfeeding 0.003
10 B11 Preceding birth interval 0.003

🚀 How to Run

# Clone
git clone https://github.com/Derio001/chad-malnutrition-prediction
cd chad-malnutrition-prediction

# Install dependencies
pip install pandas numpy matplotlib seaborn scikit-learn
pip install xgboost catboost

# Run notebook
jupyter notebook Chad_Malnutrition_Prediction.ipynb

Note on data: DHS data requires free registration at dhsprogram.com. Request Chad 2014 KR recode dataset. The processed modeling dataset is included in this repo.


📁 Project Structure

chad-malnutrition-prediction/
│
├── Chad_Malnutrition_Prediction.ipynb    # Full pipeline
├── chad_malnutrition_model_data.csv      # Processed dataset
├── malnutrition_model_comparison.csv     # Model results
├── malnutrition_feature_importance.csv   # Feature rankings
└── README.md

🔭 Future Work

  • Regional disaggregation — Sahel vs southern Chad
  • Integrate with crop yield early warning system
  • Add 2023 DHS data when available for Chad
  • Build Streamlit dashboard for field worker use
  • Extend to severe acute malnutrition (SAM) prediction
  • Multi-country model (Niger, Mali, Sudan)

🏛️ Policy Implications

Organization Application
UNICEF Chad Target child nutrition programs
WFP Chad Link food security to malnutrition risk
MSF / Action Against Hunger Field screening tool
Ministry of Health Chad National nutrition surveillance
World Bank Evidence base for nutrition investment

👤 Author

Mahamat Hanga Derio M.Tech Data Science — Christ University, Bangalore Chadian national | Building ML tools for public health and development in Sub-Saharan Africa

📬 Open to collaboration with UNICEF, WFP, WHO, NGOs and research institutions working on child health in Chad 🔗 GitHub | LRI Project | Crop Early Warning | GDP Dashboard


Part of an integrated data science portfolio focused on child welfare, food security, and economic development in Chad and the Lake Chad Basin.


---

Now create the GitHub repo `chad-malnutrition-prediction`, upload the files, paste this README, and add topics:

machine-learning chad malnutrition child-health dhs-data africa unicef public-health python gradient-boosting xgboost

About

ML pipeline predicting child malnutrition risk in Chad using DHS 2014 survey data. Gradient Boosting achieved 92% accuracy and 0.979 AUC on 9,826 children. 52.9% of Chadian children under five are malnourished.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Contributors

Languages