jessicanigre-droid/Explainable-Machine-Learning-for-Childhood-Malnutrition-in-Ghana-DHS-2022-

Compares logistic regression, LASSO, Random Forest, and XGBoost to predict stunting, wasting, and underweight in Ghanaian children (DHS 2022). Using MICE, SMOTE, and SHAP, XGBoost led with AUC 0.912–0.982. Key predictors: age, birth size, anaemia, maternal education, antenatal visits, and household environment. ML outperforms traditional methods. # Explainable-Machine-Learning-for-Childhood-Malnutrition-in-Ghana-DHS-2022- Compares logistic regression, LASSO, Random Forest, and XGBoost to predict stunting, wasting, and underweight in Ghanaian children (DHS 2022). Using MICE, SMOTE, and SHAP, XGBoost led with AUC 0.912–0.982. Key predictors: age, birth size, anaemia, maternal education, antenatal visits, and household environment. ML outperforms traditional methods. # Malnutrition Prediction with ML (GDHS 2022) ## Overview This project compares **logistic regression**, **LASSO**, **Random Forest**, and **XGBoost** to predict stunting, wasting, and underweight among Ghanaian children under five using the **2022 Ghana Demographic and Health Survey (GDHS)**. ## Methods - **MICE**: handled missing data (breastfeeding/complementary feeding gaps) - **SMOTE**: corrected class imbalance in training sets - **SHAP**: explained model predictions and identified key drivers ## Results - **XGBoost** achieved the best performance (AUC 0.912–0.982) - **Logistic regression** performed weakest (AUC ~0.67–0.70) - **Key predictors**: child age, birth size, anaemia status, maternal education, antenatal visits, household environment (water, sanitation, wealth) ## Impact Machine learning models outperform traditional approaches, offering **actionable insights** for policy makers and targeted interventions to reduce childhood malnutrition in Ghana. ## Repository Structure - `data/` – GDHS 2022 dataset (not included, request via DHS Program) - `notebooks/` – preprocessing, imputation, and model training scripts - `results/` – performance metrics, SHAP plots, and risk profiles - `README.md` – project overview