If you are talking about ThingWorx Analytics prediction models, then there is indeed not an easy way to know how much a model training will take.
Here are a few points that do impact the training time:
As you indicated the size of the dataset impacts .
Note that when we say size, it includes both the number of row and the number of columns in the dataset.
It is important to clean the dataset first to make sure to not include necessary columns.
If using categorical dataset, the number of category will impact a lot the performance too, so try to keep it to a minimal
The number of learners selected will also have an impact.
The article https://www.ptc.com/en/support/article/CS248206 has also some info but there is no formula that I know that can tell how much a model will take