Predicting children’s undernutrition and its association with women’s asset ownership in Ethiopia: spatial machine learning

Abstract

Child undernutrition remains a critical public health concern, with prominent geographic and structural disparities in Ethiopia. Understanding how women's asset share and spatial context interact to affect nutritional risk is vital for effective interventions. This study investigates the spatial patterns of child undernutrition, its association with women's asset ownership and other covariates, and the predictive performance of spatial machine learning techniques in making predictions. Using the nationally representative ESPS 2021/22 for model development and internal validation through spatial cross-validation with a 70/30 split, alongside temporal validation employing ESPS I, 2018/19. The metrics were used to compare four machine learning algorithms (random forest, gradient-boosted trees, support vector machines, and geographical random forests) with spatial features to a spatial generalized linear mixed model. The semivariogram result showed a strong spatial dependence, which supported the use of spatial modeling. The women's asset ownership index, children's age, distance to nearest market, mothers' age, distance to nearest road, household size, child sex, region, toilet type, mother's religion, source of drinking water, and place of residence were significant factors. The results demonstrate a distinct inverse spatial correlation between women's asset ownership and child undernutrition, suggesting that women's economic empowerment functions as a protective structural part. The result of temporal validation showed a growing impact of geographic and contextual factors. Gradient-boosted trees and random forests with spatial features make it easy to model intricate nutritional outcomes in different settings. The results underline geographically targeted child nutrition programs and policies that bolster women's economic autonomy and resilience. Integrating spatial analytics and Gradient-boosted trees and random forests machine learning techniques with gender‑sensitive socioeconomic factors can strengthen precise public health strategies to reduce child undernutrition.