Also, saying "subset data sets" is redundant. The word "subsets" already includes "sets", and basically means "a smaller set from a larger set." So, I would describe what you are extracting as data subsets, as in the following examples:
Given the original data, we extracted various subsets based on the
We extracted these data subsets from the original study...
We used random sampling to extract a subset of data...
All of them are correct, except that get a new... seems more natural than get new.... Also note that assemblies derived via random sampling are often referred to as drawn rather than created (e.g.: "Eight sets/subsets/datasets/samples were drawn from the population") to underpin the notion they are random rather than deterministically constructed.
Are you sure you mean title? Because it seems you talk about an abstract. A title would sound like "Creation of a subset...".
Indeed, extract is commonly used in sampling, and you can also simply say we sample a subset or we take/draw a sample Get is informal and is not preferable in formal literature. Construct has a nuance of elaboration, often as a result of an algorithm or procedure in which all steps are given. Obtain is very common and has a formal touch.
When speaking precisely about any type of set, a given set is also defined as a "Subset" of itself.
If you want to precisely describe a partial dataset, you would instead refer to a "Proper Subset" — so perhaps one might say you will be "selecting and/or extracting (proper) subsets from the entire dataset."
But omitting "proper" is perfectly fine, it's just if you need to specifically emphasize the non-entirety of it for some reason.