Thanks @jcackler, appreciate the reply! I will definitely check out the Temporal Transformer, if not for anything else than to learn something :D As for the Tabular Classifier, and I'm not sure if the formatting of your post is messed up, but am I correct in assuming that you're saying that (for example) the last 10 samples are refactored into a single data row (or sequence, if you will)? Yeah, I think that may be worth a look as well. I think I've managed to discard a few features, so I'm down to 8 where the timestamp/packet id is one, but in this case I think it may well be discarded. However, let's say for the sake of argument that I need to examine the last 10 seconds @ 60Hz, that would be (target state + 7 features times 600) 4201 entries per row. Will the sheer amount of data cause any issues? Anyway, I will look into these approaches!