Bayesian Network Analysis of Determinants of Child Stunting in North East Nigeria

Child stunting, defined as height-for-age z-score below minus two standard deviations from the WHO median, affects approximately 52 percent of children under five in North East Nigeria, reflecting compounded effects of nutritional deprivation, recurrent illness, and inadequate childcare practices. This study applies Bayesian network structure learning and probabilistic inference to identify the causal pathways and conditional dependencies among determinants of stunting in a sample of 2,800 children under five from Adamawa, Gombe, and Yobe States, using data from the 2021 National Nutrition Survey. Bayesian network structure is learned from data using the PC algorithm with constraint-based testing and subsequently refined using expert knowledge encoding. Variables in the network include maternal height, minimum dietary diversity, exclusive breastfeeding duration, birth weight, water source safety, sanitation access, household food insecurity score, and stunting outcome. Probabilistic inference through belief propagation identifies the conditional probability of stunting given combinations of risk factor status. Results reveal that the joint presence of low dietary diversity and unsafe water source elevates stunting probability to 0.74, compared to 0.31 in the presence of neither risk factor. Maternal height has the second largest marginal influence on stunting probability. The study recommends prioritising complementary feeding improvement and water safety interventions in integrated nutrition programming for the North East. Keywords: Bayesian network, child stunting, North East Nigeria, nutritional determinants, probabilistic inference.