Is this question about categorical features on-topic?

Question

The question Suggestions on using model in production 1 test at a time was closed as off-topic.

To my mind, the question content is squarely about the generalizability of a specific kind of feature (categorical encodings of particular strings). This seems on-topic! The subject matter seems incidental because OP's question is about a more general kind of dummy variable trap, in the same way that you wouldn't use a categorical encoding of SSN or phone number in a regression model because it uniquely identifies each respondent.

I think that my answer directly addresses the statistical component in its first section. The second section of my answer is probably not on-topic, but since I happen to know about these models, I felt that I could offer some constructive advice. The third section of my answer is about how to construct generalizable features or else use feature-free architectures which support generalizable models.

Would anyone involved in the closure of this question be willing to elaborate on their decision to close?

Answer 1

I voted to close as "Unclear what you're asking" because I genuinely could not follow the question. The main statement of the problem is:

My big question is how do I handle one test feature where part of it has not been trained?

...which I didn't really understand till I read your answer just now. And an apparent secondary problem was:

The issue that I am running into is the training set has seen the ChildPath "C:\Windows\System" and the ChildExe "cmd.exe" which are normal, but the training set has not seen the ParentPath "C:\Windows\Malicous" or ParentExe "badscipt.exe" so these have not been label or one hot encoded.

...which seems like more of a programming issue.

I notice that your answer also begins:

The challenge that you've discovered is a natural consequence of how you've organized your research project: your model has no generalizable information about new file paths or new names of .exe files.

This also seems to suggest a programming/organisational problem rather than a statistical/machine learning one.

That said, I don't have much expertise in this. So my confidence in my judgement is not very high and I'm willing to reverse it. I do think the question is poorly written, though, and would benefit from some editing.