Predicting severe stunting and its determinants among under-five in Eastern African Countries: A machine learning...

Area of Science:

• Public Health
• Pediatrics
• Machine Learning Applications

Background:

• Severe stunting is a critical public health issue in low- and middle-income countries, particularly in Eastern Africa, impacting millions of children under five and contributing to mortality.
• Limited research has explored severe stunting using machine learning (ML) in Eastern Africa, highlighting a gap in understanding its determinants.

Purpose of the Study:

• To predict severe stunting among children under five in Eastern Africa using ML algorithms.
• To identify the major determinants contributing to severe stunting.
• To enhance model interpretability using Shapley Additive explanations (SHAP) and Association Rule Mining (ARM).

Main Methods:

• A cross-sectional study utilizing Demographic and Health Survey (DHS) data from 2012-2022 in Eastern Africa.
• Analysis of data from 76,019 children, sourced from 136,074 children, using Python and R.
• Model performance evaluated using accuracy and Area Under the Curve (AUC), with SHAP and ARM for determinant interpretation.

Main Results:

• The Random Forest model demonstrated the highest performance with 87% accuracy and an AUC of 0.83.
• Key predictors for increased severe stunting risk included not practicing exclusive breastfeeding, being from Burundi, child underweight status, poor household conditions, male gender, short maternal height, maternal underweight, small birth size, home delivery, primary maternal education, unimproved toilet facilities, and distance to health facilities.

Conclusions:

• The Random Forest model effectively predicts severe stunting in Eastern African populations.
• Integrated interventions are crucial, focusing on socioeconomic support for mothers, enhanced maternal education, promotion of exclusive breastfeeding and facility deliveries, improved sanitation and hygiene, and accessible healthcare services.