Predicting Child Malnutrition in Pakistan Using Machine Learning: Evidence from PDHS

Child stunting, measured by height-for-age Z-score (HAZ), remains a major public health issue in developing countries, including Pakistan. Identifying its key determinants is essential for effective policy formulation and intervention strategies. This study uses data from the Pakistan Demographic and Health Survey (PDHS) 2017–18. The dependent variable, height-for-age Z-score (HW70), was converted into a binary indicator of stunting. Independent variables include child characteristics (sex, age, birth order), maternal factors (age, education, antenatal care), household conditions (wealth index, residence, region, water, sanitation, and media access), and health indicators (breastfeeding duration, postnatal care, and recent illnesses). Four-machine learning models Logistic Regression, K-Nearest Neighbors (KNN), Random Forest, and Support Vector Machine (SVM) were applied. Model performance was evaluated using accuracy, precision, recall, F1-score, area under the ROC curve (AUC), and mean squared error (MSE). The findings indicate that the Random Forest model outperforms other models across most evaluation metrics, particularly in terms of F1-score and AUC, demonstrating strong predictive capability. Logistic Regression shows competitive performance, especially in accuracy and precision, while KNN and SVM exhibit relatively lower performance. The study highlights the effectiveness of ensemble learning techniques in predicting child stunting and underscores the importance of socioeconomic, maternal, and health-related factors. These findings can support policymakers in designing targeted interventions to reduce child malnutrition in Pakistan.