[zelphirkalt]

I think people throwing algorithms at problems at Kaggle or combining them is not the kind of research, which the GP was referring to.

Limiting a tree by its depth is a very general global parameter for a tree. One could try to use any kind of criteria for deciding when to stop making more child nodes in a tree, depending on what information is locally available and that depends on how the tree algorithm is actually implemented. So people doing Kaggle challenges would have to dig into the source code of the tree implementation, then change things there, to modify locally available knowledge, in order to allow for more fine grained decision making at each node.

That is only the constructive side of things, when the tree is created. Even more powerful is the post processing / destructive / prunning side of things, because theoretically the whole tree structure can be taken into account, when deciding what branch to cut.

I think the GP is referring to research in the area of what other useful things one can come up with here. As far as I know, these are not the usual things people do in Kaggle challenges. Correct me if I am wrong.