Child undernutrition remains a critical public health challenge in Pakistan. This study evaluated five machine learning algorithms logistic regression, linear discriminant analysis, k-nearest neighbors, support vector machines (SVM), and random forest (RF) to predict undernutrition among children under five using Pakistan Demographic and Health Survey 2017–18 data (n=4,098). Undernutrition prevalence was 44.2% using the Composite Index of Anthropometric Failure. Random Forest achieved superior predictive performance (AUC: 0.86, accuracy: 80.0%, F1: 0.85), outperforming SVM (0.81), logistic regression (0.79), LDA (0.78), and KNN (0.75). Additionally, RF-based feature importance found that the maternal BMI, household wealth, sanitation, and child age were key predictors. These findings confirm that random forest outperforms other algorithms, supporting its integration into health information systems for targeted interventions.

Prediction, undernutrition, machine learning, Feature importance, Pakistan

(1) Muhammad Ali Yahya
Bachelors of Artificial Intelligence, Department of Artificial Intelligence, School of Computing, The Islamia University of Bahawalpur, Bahawalpur, Punjab, Pakistan.
(2) Ahmad Samama Salehien
Bachelors of Computer Science, Department of Computer Science, School of Computing, The Islamia University of Bahawalpur, Bahawalpur, Punjab, Pakistan.
(3) Shahzad Jahangir
PhD Scholar (Development Studies Major in Health Economics), PIDE School of Policy, Development and Governance (PSPDG), Islamabad, Pakistan.