Smart handeling of pipeline timeouts

I am currently working on a problem where i'm optimizing a pipeline quite a large dataset (100,000 obs, 80 features). A lot of the pipelines (approx 20/100) time out with the time limit on 15 minutes, which is basically a waste of computation time. Would there be any way to avoid such cases?

Possibilities:
- Keep counts of timeout of each operator, and disable a operator during the optimization when it has caused a number of timeouts.
- Limit the 'config' a priori, depending on the timeout parameters, based on rules of thump or metamodels.