Hybridizing spatial machine learning to explore the fine-scale heterogeneity between stunting prevalence and its associated risk determinants in Rwanda
Nduwayezu, Gilbert LU
; Mansourian, Ali
LU
; Bizimana, Jean Pierre
and Pilesjö, Petter
LU
(2025)
In Geo-Spatial Information Science
- Abstract
Childhood stunting is a serious global public health issue that exhibits local spatial variations. Previous studies have used traditional statistical methods to identify stunting risk factors, and little is known about the application and usefulness of spatial machine learning techniques in identifying localized stunting risk factors based on complex datasets. This study assesses the performance of hybrid spatial machine learning techniques in predicting stunting among children below the age of five in Rwanda. We cross-sectionally examined Bayesian-modeled surface stunting prevalence data linked with their related covariates obtained from the 2019–2020 Rwanda Demographic and Health Survey. Using these datasets, we implemented... (More)
Childhood stunting is a serious global public health issue that exhibits local spatial variations. Previous studies have used traditional statistical methods to identify stunting risk factors, and little is known about the application and usefulness of spatial machine learning techniques in identifying localized stunting risk factors based on complex datasets. This study assesses the performance of hybrid spatial machine learning techniques in predicting stunting among children below the age of five in Rwanda. We cross-sectionally examined Bayesian-modeled surface stunting prevalence data linked with their related covariates obtained from the 2019–2020 Rwanda Demographic and Health Survey. Using these datasets, we implemented geographical weighted summary statistics, global random forest, and hybrid random forest model complimented with interpretable machine learning to identify local disparities in the association between stunting prevalence and its related risk factors. The results revealed significant variation in stunting prevalence within different areas, with the Western and Northern Province regions exhibiting higher stunting prevalence compared to the other provinces in the country. Our findings demonstrate the superiority of the hybrid random forest model over the global random forest model in achieving a more accurate fit when explaining stunting prevalence. Additionally, our findings reveal a non-linear relationship between stunting prevalence risk and its predictors. Specifically, we observed the highest risk of stunting when the percentage of households without toilet facility reached 2%. However, when the proportion of antenatal visits, men’s education, women’s literacy, access to clean water, and delivery place reached 50%, 85%, 80%, 70%, and 95%, respectively, the risk of stunting prevalence was at its lowest point. Furthermore, our findings indicate a lower prevalence of stunting when less than 20% of households use insecticide-treated nets. Localized information on stunting is highly valued by stakeholders for measuring and monitoring progress toward sustainable development goals.
(Less)
- author
- Nduwayezu, Gilbert
LU
; Mansourian, Ali
LU
; Bizimana, Jean Pierre
and Pilesjö, Petter
LU
- organization
- publishing date
- 2025
- type
- Contribution to journal
- publication status
- epub
- subject
- keywords
- hybrid random forest, interpretable machine learning, local variation, nonlinear effects, Stunting prevalence
- in
- Geo-Spatial Information Science
- publisher
- Taylor & Francis
- external identifiers
-
- scopus:86000522796
- ISSN
- 1009-5020
- DOI
- 10.1080/10095020.2025.2459133
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2025 Wuhan University. Published by Informa UK Limited, trading as Taylor & Francis Group.
- id
- ad5fa9d9-34e4-4c91-8e40-25c7f87d42ee
- date added to LUP
- 2025-03-27 22:44:25
- date last changed
- 2025-04-04 14:04:32
@article{ad5fa9d9-34e4-4c91-8e40-25c7f87d42ee,
abstract = {{<p>Childhood stunting is a serious global public health issue that exhibits local spatial variations. Previous studies have used traditional statistical methods to identify stunting risk factors, and little is known about the application and usefulness of spatial machine learning techniques in identifying localized stunting risk factors based on complex datasets. This study assesses the performance of hybrid spatial machine learning techniques in predicting stunting among children below the age of five in Rwanda. We cross-sectionally examined Bayesian-modeled surface stunting prevalence data linked with their related covariates obtained from the 2019–2020 Rwanda Demographic and Health Survey. Using these datasets, we implemented geographical weighted summary statistics, global random forest, and hybrid random forest model complimented with interpretable machine learning to identify local disparities in the association between stunting prevalence and its related risk factors. The results revealed significant variation in stunting prevalence within different areas, with the Western and Northern Province regions exhibiting higher stunting prevalence compared to the other provinces in the country. Our findings demonstrate the superiority of the hybrid random forest model over the global random forest model in achieving a more accurate fit when explaining stunting prevalence. Additionally, our findings reveal a non-linear relationship between stunting prevalence risk and its predictors. Specifically, we observed the highest risk of stunting when the percentage of households without toilet facility reached 2%. However, when the proportion of antenatal visits, men’s education, women’s literacy, access to clean water, and delivery place reached 50%, 85%, 80%, 70%, and 95%, respectively, the risk of stunting prevalence was at its lowest point. Furthermore, our findings indicate a lower prevalence of stunting when less than 20% of households use insecticide-treated nets. Localized information on stunting is highly valued by stakeholders for measuring and monitoring progress toward sustainable development goals.</p>}},
author = {{Nduwayezu, Gilbert and Mansourian, Ali and Bizimana, Jean Pierre and Pilesjö, Petter}},
issn = {{1009-5020}},
keywords = {{hybrid random forest; interpretable machine learning; local variation; nonlinear effects; Stunting prevalence}},
language = {{eng}},
publisher = {{Taylor & Francis}},
series = {{Geo-Spatial Information Science}},
title = {{Hybridizing spatial machine learning to explore the fine-scale heterogeneity between stunting prevalence and its associated risk determinants in Rwanda}},
url = {{http://dx.doi.org/10.1080/10095020.2025.2459133}},
doi = {{10.1080/10095020.2025.2459133}},
year = {{2025}},
}