Artificial Intelligence for Child Stunting Prediction in India: A Brief Narrative Review on Models, Performance, and Public Health Applications

Background: Childhood stunting is a serious public health issue in India, where it affects 35.5% of children below 5 years.There has been substantial growth in applying artificial intelligence (AI) and ML models to large population data.Objectives: The purpose of this brief narrative review was to examine the literature on the use of AI algorithms to predict stunting among Indian children and other South Asian populations, focusing on AI models' architecture, predictive performance, determinants, and applications.Materials and methods: Six articles published from 2017 to 2024 were selected.The studies applied ML and advanced statistics techniques using datasets such as National Family Health Survey (NFHS) and Demographic and Health Survey (DHS).The analysis of the data was carried out through various dimensions such as models used, outcome variables, predictors, and performance measures.Findings: Models using ensemble learning techniques such as random forest and gradient boosting achieved the highest accuracy in comparison to conventional logistic regression models.The accuracies ranged from 62 to 96%, while AUC-ROC scores varied from 0.66 to 0.99 based on outcomes considered.Common predictors used in AI models were the nutritional status of mothers, the age of children, the socioeconomic status of households, and antenatal care.Although AI models displayed potential in capturing interactions, limitations such as the use of crosssectional data and class imbalance constrained performance.Conclusion: Artificial intelligence models offer an exciting technology that could be effectively utilized in identifying and stratifying risks of child undernutrition in India.Nevertheless, several methodological and operational issues have to be considered.