Sorry for not replying earlier, somehow I missed this comment.

For me, the question is what you are going to do with that afterwards? If you want to use that to choose between different forecasting techniques, maybe just a simple split to “have zeroes/doesn’t have” would suffice. If it is just a curiosity, then who cares how you categorise the demand?

Also, have a look at this: https://openforecast.org/2025/04/11/svetunkov-sroginis-2025-model-based-demand-classification/ – I’ve been working with Anna Sroginis on an alternative classification. The main idea is to first understand whether the zeroes happen naturally or artificially and then do a classification to regular/intermittent (with lumpy/smooth inside). This paper is under review right now, when it gets published, I’ll write more about it (it has already changed a bit).