Machine Learning Prediction of Childhood Anaemia and Its Association with Household Livestock Ownership in Uganda: Evidence from the 2016 Demographic and Health Survey

Childhood anaemia and household livestock ownership may be connected through nutritional, economic and infection-related pathways, making both prediction and association analysis relevant to Uganda. This article presents an integrative literature review and a prespecified protocol for analysing the 2016 Uganda Demographic and Health Survey. Twenty-six sources were retained through targeted scientific and institutional searches. The official survey report records anaemia in 52.8% of tested children aged 6–59 months, with pronounced age variation. Published livestock studies show mixed and species-specific associations, while a comparable Ethiopian prediction study found that logistic regression outperformed the other algorithms it assessed. These findings support evaluating model performance empirically rather than presuming that ensemble methods will be superior or livestock ownership protective. The proposed analysis will use eligible household-member records, verified livestock variables, survey design information and haemoglobin-based outcomes. A historical DHS definition will be primary, with a separately specified sensitivity analysis for the 2024 World Health Organization guidance. Survey-weighted association models will be distinguished from predictive models. Prediction will compare a prevalence-only benchmark, penalised logistic regression, random forest and gradient boosting using nested validation grouped by survey cluster. Preprocessing will remain inside training folds; haemoglobin and outcome-derived variables will be excluded from predictors. Evaluation will address discrimination, calibration, threshold-dependent performance, subgroup variation and the incremental contribution of livestock variables. No individual-level DHS data were analysed for this article, and no original livestock effect or model accuracy is reported. The protocol provides a reproducible basis for subsequent approved analysis while identifying limits to causal interpretation and contemporary deployment.