Disaggregating Census Data for Population Mapping Using a Bayesian Additive Regression Tree Model

Fine-scale population census data are often lacking due to the challenge of sharing such sensitive data at granular scales. In this study, we compare the Random Forest (RF) model and the Bayesian Additive Regression Tree (BART) model for population disaggregation using both census data from Ghana and simulated data. The BART model outperforms the RF model in out-of-sample predictions for metrics like bias, mean squared error, and root mean squared error. It also provides uncertainty estimates around the predicted population, which is often lacking with the RF model. This study highlights the BART model's superiority in disaggregating population data.