Modern Prediction Models for BMI: Contrasting XGBoost with Bayesian Geo-additive Quantile Regression Model in Women's Health

Estimating body mass index (BMI) outcomes among women of reproductive age helps inform public health policies and especially in nutritional epidemiology. This study compares two advanced modeling approaches—Extreme Gradient Boosting (XGBoost) and a Bayesian Geo-Additive Regression Model—using data from the 2018 Nigerian Demographic and Health Survey (NDHS). The NDHS employed a two-stage stratified random sampling design which helps provide a representative sample across urban and rural areas. Model performance is assessed using key metrics including accuracy, root mean squared error (RMSE), and area under the receiver operating characteristic curve (AUC-ROC). Uncertainty estimation is achieved by comparing Bayesian credible intervals with bootstrapped prediction variability from XGBoost. Also, we evaluated intepretability considering feature importance metrics. Our findings show that XGBoost offers higher computational efficiency and marginally better predictive accuracy, the Bayesian geo additive Regression model gives more robust estimates of uncertainty and thereby enhanced interpretation, which is very important for risk assessment in public health. These findings underscore the trade-offs between computational speed and probabilistic insight hinting towards a hybrid modeling frameworks as further improvement on predictive performance.