Peer-reviewed article
Machine Learning in Predicting Child Malnutrition: A Meta-Analysis of Demographic and Health Surveys Data.
Bhagyajyothi Rao, Muhammad Rashid, Md Gulzarull Hasan, Girish Thunga ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This meta-analysis reviewed 11 peer-reviewed studies published between 2019 and 2023 that used machine learning to predict child malnutrition using Demographic and Health Survey data. Most studies came from Bangladesh and focused on stunting, wasting, and underweight in children under five. Machine learning models showed moderate to good predictive accuracy, with stunting predicted at 69%, wasting at 84%, and underweight at 74% accuracy across the pooled results.
Key points
- Of 789 studies screened, 11 met inclusion criteria for machine learning prediction of malnutrition in children under 5 using DHS data
- Six of 11 studies (55%) were conducted in Bangladesh
- Pooled accuracy for predicting stunting was 68.92% across 10 studies
- Pooled accuracy for predicting wasting was 84.39% across 3 studies
- Pooled accuracy for predicting underweight was 73.60% across 4 studies
Limitations
The meta-analysis does not explain why machine learning models would be used instead of simpler direct measurement approaches for malnutrition assessment, nor does it clarify the practical implementation context or generalisability of these models beyond the DHS datasets studied. The high heterogeneity (I² = 100%) suggests findings may not be directly comparable across studies.
Abstract
BACKGROUND: Childhood malnutrition remains a significant global public health concern. The Demographic and Health Surveys (DHS) program provides specific data on child health across numerous countries. This meta-analysis aims to comprehensively assess machine learning (ML) applications in DHS data to predict malnutrition in children. METHODS: A comprehensive search of the peer-reviewed literature in PubMed, Embase, and Scopus databases was conducted in January 2024. Studies employing ML algorithms on DHS data to predict malnutrition in children under 5 years were included. Using PROBAST (Prediction model Risk Of Bias Assessment Tool), the quality of the listed studies was evaluated. To conduct meta-analyses, Review Manager 5.4 was used. RESULTS: A total of 11 out of 789 studies were included in this review. The studies were published between 2019 and 2023, with the major contribution from Bangladesh (n = 6, 55%). Of these, ten studies reported stunting, three reported wasting, and four reported underweight. A meta-analysis of ten studies reported a pooled accuracy of 68.92% (95% CI: 66.04, 71.80; I2 = 100%) among ML models for predicting stunting in children. Three studies indicated a pooled accuracy of 84.39% (95% CI: 80.90, 87.87; I2 = 100%) in predicting wasting. A meta-analysis of four studies indicated a pooled accuracy of 73.60% (95% CI: 70.01, 77.20; I2 = 100%) for ML models predicting underweight status in children. CONCLUSIONS: This meta-analysis indicated that ML models were observed to have moderate to good performance metrics in predicting malnutrition using DHS data among children under five years.