Journal of Genomics and Precision Medicine

ABSTRACT

The intricate interplay between social, environmental, and biological variables that cause child- hood malnutrition in low- and middle-income nations remains a global public health issue. In health survey data, statistical and machine learning methods generally fail to account for genetic susceptibility, nonlinear interactions, uncertainty quantification, and measurement error. This paper presents a hybrid Bayesian machine learning paradigm for genomic risk prediction in childhood malnutrition using demographic, environmental, and genetic data. Combining Bayesian hierarchical measurement error modeling, Bayesian Deep Neural Networks (BDNN), Graph Neural Networks (GNN), and Explainable Artificial Intelligence (XAI) improves predicted accuracy and interpretability. Genomic predictors, including polygenic risk scores, were combined with demographic and health survey data. Bayesian inference and uncertainty quantification were used with Hamiltonian Monte Carlo estimation and probabilistic deep learning. The Bayesian AI framework outperforms statistics and machine learning techniques. The Graph Neural Network had the greatest prediction performance with an AUC-ROC of 0.968 and 93.7% classification accuracy. Explainable AI analysis found that polygenic risk scores, maternal education, family affluence, and sanitation access best predict childhood malnutrition. Bayesian measurement error correction enhanced model resilience and probabilistic calibration. The suggested precision health framework for early malnutrition risk diagnosis, genetic epidemiology, and AI-driven nutritional monitoring is strong. Bayesian inference, machine learning, and genetic analytics provide tailored dietary treatments and evidence-based public health decisions.

Keywords: Childhood Malnutrition; Bayesian Machine Learning; Genomic Risk Pre- diction; Precision Nutrition; Explainable Artificial Intelligence (XAI)