Bayesian additive regression trees for imbalanced ordinal outcomes

The ordered probit Bayesian additive regression trees (OPBART) model is an extension of the Bayesian additive regression trees (BART) framework designed to handle ordinal response variables.In this study, we apply OPBART to imbalanced ordinal data.Specifically, we employ OPBART to predict smoking level using real-world data from the 2023 Korea National Health and Nutrition Examination Survey (KNHANES 2023), in which the response variable (smoking level) exhibits a highly imbalanced distribution.To examine more concretely how well the OPBART handles imbalanced data, we conducted four simulation scenarios under both balanced and imbalanced conditions.We compare the predictive performance of OPBART with alternative models in both the real data analysis and simulation study to determine which method performs best.The results demonstrate that OPBART outperforms competing models.These findings suggest that OPBART can serve as a promising alternative for modeling imbalanced ordinal outcomes.