 This paper investigated the accuracy of three popular tree-based machine learning models, chi-square automatic interaction detection, CHAID, random forest, RF, and gradient-boosted tree, GBT, when used to predict bus travel times on high and low-frequency routes. The authors found that CHAID was the most accurate model for predicting travel times on high-frequency routes, while GBT performed better than RF on low-frequency routes. Additionally, they developed a new stop-based root construction method that proved to be more accurate than existing methods. This research provides valuable insights into how to accurately predict bus travel times on different types of routes, and could help improve the efficiency of public transit systems.