Predicting risk of inadequate micronutrient intake with transferable machine learning models.

Identifying populations at risk of inadequate micronutrient intake is necessary for governments and development partners in low- and middle-income countries to make informed and timely decisions on nutrition-relevant policies and programmes. In this study, we propose a machine learning methodological approach using data on household dietary diversity, socioeconomic status, and climate indicators to predict the risk of inadequate micronutrient intake. Using case studies from Ethiopia and Nigeria, we demonstrate that the models effectively predict risk, with key predictors showing consistency in terms of importance and direction. We also illustrate the feasibility of transferring models between countries, offering a short-term, practical solution for contexts lacking nationally representative micronutrient data. Our results show that this machine learning methodological approach can generate geographically and socioeconomically disaggregated risk estimates that reflect expected patterns of nutritional vulnerability, supporting more targeted and data-driven nutrition interventions.