According to Butina et al. (2003), where CUBIST was compared to another method, the exact algorithm behind the method is proprietary to the author (Quinlan), but there is an early publication available:

• Quinlan, J. R. (1992, November). Learning with continuous classes. In Proceedings of the 5th Australian joint Conference on Artificial Intelligence (Vol. 92, pp. 343-348). http://sci2s.ugr.es/keel/pdf/algorithm/congreso/1992-Quinlan-AI.pdf

And here the website of RuleQuest, the company founded by Quinlan that seems to distribute tools using the method:

• https://www.rulequest.com/

They also have quite a substantial list of publications that used CUBIST:

• https://www.rulequest.com/cubist-pubs.html

This site about the {caret} R package, from the same author (Max Kuhn) as {cubist} might also be helpful:

• http://topepo.github.io/caret/index.html

Or the following slides from a presentation of Kuhn about how to use {caret} for predictive modeling:

• http://www.edii.uclm.es/~useR-2013/Tutorials/kuhn/user_caret_2up.pdf

References:

• Butina, D., & Gola, J. M. (2003). Modeling aqueous solubility. Journal of chemical information and computer sciences, 43(3), 837-841.

Thank you very much AB