Integrating machine learning with Bayesian structured additive spatio-temporal modeling to examine inequalities in antenatal care utilization in Nigeria

Title : Integrating machine learning with Bayesian structured additive spatio-temporal modeling to examine inequalities in antenatal care utilization in Nigeria

Abstract:

Improving access to adequate antenatal care (ANC) remains a major global health priority for reducing maternal and neonatal mortality. Despite considerable progress in maternal health programs, substantial disparities in the utilization of recommended antenatal services persist across many low and middle-income countries. In Nigeria, the uptake of the recommended four or more antenatal visits (ANC4+) remains uneven across socioeconomic groups, communities, and geographic regions, posing a significant challenge to improving maternal health outcomes and equitable access to maternal healthcare services. This study proposes a hybrid analytical framework that integrates machine learning with Bayesian structured additive spatio-temporal modeling to investigate determinants and geographic patterns of optimal antenatal care utilization in Nigeria. Data are drawn from multiple waves of the nationally representative Nigeria Demographic and Health Survey (NDHS). In the first stage, Extreme Gradient Boosting (XGBoost) is applied to capture complex nonlinear relationships and interactions among socioeconomic and demographic predictors of ANC utilization. Predicted probabilities from the machine learning model are used to generate an Artificial Intelligence (AI) risk score summarizing latent nonlinear effects and predictive patterns in the data. In the second stage, the AI-derived risk score is incorporated into a Bayesian Structured Additive Regression (STAR) model estimated using Integrated Nested Laplace Approximation (INLA). This framework simultaneously models fixed covariate effects, nonlinear predictor effects, spatially structured and unstructured regional variation, temporal trends across survey waves, and space-time interaction effects. The analysis is expected to identify key socioeconomic determinants and geographic clusters associated with low ANC4+ utilization while revealing regions where disparities persist or evolve over time. To our knowledge, this study is among the first to integrate machine learning with Bayesian structured additive spatio-temporal modeling to examine antenatal care utilization in Nigeria using nationally representative survey data.
Keywords: Antenatal Care; Machine Learning; Bayesian Structured Additive Regression; Spatio-Temporal Analysis; Maternal Health; Nigeria.

Biography:

Paul Omoh Olopha is a researcher and academic affiliated with the Federal University of Technology, Akure, Nigeria. His research focuses on advanced statistical modeling, spatial epidemiology, and the application of machine learning techniques in public health and demographic research. He specializes in analyzing maternal and child health outcomes, health inequalities, and population health dynamics in low and middle-income countries. His work frequently utilizes large-scale population datasets, including the Nigeria Demographic and Health Survey, and employs Bayesian and artificial intelligence-driven approaches to investigate complex health patterns and generate evidence to support data-driven public health policy and intervention planning.