It sounds like a rename (via the moderator merge tool) of classification to image-classification would be a quick and easy fix to a large part of the issue identified.
After/before that we just need to disambiguate any questions that are (to be) tagged image-classification and are not about image classification.
One set of questions where classification is used, I think incorrectly, can be identified using https://gis.stackexchange.com/search?q=%5Bclassification%5D+symbology+is%3Aquestion. These questions are about symbol categories and I propose that they should be tagged with symbology and have classification removed.
This answer makes no comment on the case for supervisedunsupervisedsupervised/unsupervised classification tags because that is an area where I have negligible background.