See the paper: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.85.7712&rep=rep1&type=pdf (as suggested in refactoring-ai/predicting-refactoring-ml#130).
Implementing it would be nice as a way to measure if it also works for our domain. However, we are trying to get away of process metrics, given that they are just too expensive to be calculated. Is it worth the effort?