AI-ASSISTED EARLY RISK PREDICTION OF CHILDHOOD MALNUTRITION USING MATERNAL, NUTRITIONAL, AND CLINICAL INDICATORS IN PAKISTAN

Abstract

Childhood malnutrition remains a major public-health challenge in Pakistan, requiring timely identification of children at elevated risk. This study examined the potential of artificial intelligence (AI)-assisted predictive modelling to identify childhood malnutrition using integrated maternal, nutritional, and clinical indicators. A quantitative, cross-sectional predictive design was adopted using secondary survey data. Multiple machine-learning algorithms, including logistic regression, decision tree, naïve Bayes, support vector machine, Random Forest, and Gradient Boosting, were evaluated using accuracy, precision, recall, F1-score, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC-ROC). Explainable AI using SHAP was applied to identify the relative contribution and direction of key predictors. The findings indicated that Gradient Boosting produced the strongest predictive performance, achieving 83.1% accuracy, an F1-score of 82.0%, and an AUC of .90. Child age, dietary diversity, maternal BMI, clinical risk indicators, and maternal education emerged as important predictors of malnutrition risk. The findings suggest that integrating multidimensional indicators can improve early identification of nutritional vulnerability compared with conventional single-factor assessment. The application of SHAP further enhanced model transparency by identifying the principal factors underlying individual risk predictions. The study contributes to AI-enabled preventive healthcare by proposing an interpretable predictive framework that can complement conventional nutritional screening and support earlier, targeted intervention in Pakistan.