Peer-reviewed article

Prevalence of malnutrition and associated factors in Chinese children and adolescents aged 3-14 years using machine learning algorithms.

Fangjieyi Zheng, Kening Chen, Xiaoqian Zhang, Qiong Wang, Zhixin Zhang et al. · 21 Jul 2025

Abstract

BACKGROUND: Child malnutrition represents a critical global public health issue and it is characterised by high prevalence and severe long-term consequences for growth and development. A better understanding of its contributory factors is essential to inform the design of targeted prevention strategies and evidence-based interventions. We aimed to estimate the prevalence of malnutrition in children and adolescents aged 3-14 years, and further to identify promising factors associated with child malnutrition using machine learning algorithms. METHODS: Thirty kindergartens and 26 schools were randomly selected from Beijing and Tangshan. Child malnutrition was defined according to WHO standards. Factors for child malnutrition were selected by Logistic regression and three ensemble learning algorithms. An open-access web platform was developed to facilitate calculating probabilities of child malnutrition. RESULTS: Total 18 503 children and adolescents were surveyed, and 10.93% (n = 2022) of them were found to be malnourished. Random forest emerged as the best model, as it carried the highest area under the receiver operating characteristic curve (AUROC) at 0.929. Under the implementation of random forest, top eight factors that formed the optimal set for child malnutrition prediction were identified, including age, frequency of fast food intake, frequency of late-night snacking, family history of diabetes, duration of breastfeeding, sedentary time, and parental body mass index. Further Logistic regression analyses confirmed the predictive significance of these individual factors. CONCLUSIONS: We have identified eight contributory factors for malnutrition in 3-14-year-old children and adolescents in Beijing and Tangshan, with their prediction performance optimal under random forest. More studies among independent populations are warranted to validate our findings.