LLM predicts child malnutrition with notable fairness disparities

A new study published on arXiv explores the potential of zero-shot Large Language Models (LLMs) for predicting child stunting in low- and middle-income countries, specifically focusing on data from Bangladesh. Researchers utilized GPT-4o mini to analyze population health survey data from 2007 to 2022, transforming various characteristics into prompt-based representations. The findings indicate that GPT-4o mini achieved accuracy comparable to a supervised baseline, with higher sensitivity in identifying stunting cases and stable performance across different survey periods. However, the study also highlighted significant fairness disparities related to residence and household wealth, suggesting caution before deploying such models in public health settings. AI

IMPACT Highlights potential for LLMs in public health prediction but cautions on fairness disparities before deployment.

RANK_REASON Academic paper on applying LLMs to a public health problem. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• Bangladesh Demographic and Health Survey
• CatalyzeX
• DagsHub
• Gotit.pub
• GPT-4o mini
• Hugging Face
• ScienceCast
• South Asia

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →