Exploring Risk Factors and Predictive Modeling of Child Malnutrition in Pakistan Using Machine Learning

Abstract

Objective

This study aims to identify risk factors and develop predictive models of child malnutrition (stunting, wasting, and underweight) in Pakistani children under five using machine learning approaches.

Study Design

This cross-sectional design utilized data from the Pakistan Demographic and Health Survey 2017–2018 (PDHS).

Methods

Logistic regression was employed to identify significant socio-demographic and health-related risk factors. Four machine learning models—Random Forest (RF), Support Vector Machine (SVM), Naïve Bayes (NB), and AdaBoost—were applied to predict malnutrition indicators, with performance evaluated based on accuracy, sensitivity, specificity, and F-measure using a two-stage validation strategy (10-fold cross-validation and 80:20 train-test split).

Results

Key risk factors identified included consanguineous marriages, lower wealth status, low maternal education, and geographic disparities in Sindh and Baluchistan. Among the machine learning models, Random Forest demonstrated the highest overall accuracy and specificity across all indicators, while SVM showed higher sensitivity for wasted and underweight children.

Conclusion

The study highlights the complex interplay of socio-demographic factors in child malnutrition and the potential of machine learning models to effectively predict these conditions, underscoring the need for targeted interventions focusing on maternal education, access to clean water and sanitation, and poverty alleviation, particularly in high-risk regions.