If we have a dataset with a few string-type factors that have a lot of one class, at what point do we decide to remove the said class?
This question just came to mind since I was practicing on a dataset with a factor that had only one instance of a second class and >1100 of the main class.
I had another that had only 5 cases where the result was different from the dominant class.
So is it detrimental to keep these types of variables in the model? Or is it more valuable to keep them, in case these rare instances actually give valuable information?