As far as I know, there aren’t any automated feature selection libraries. In the original paper by the authors when they built the “data science machine”, they used a few different methods for feature selection but nothing automated. In an actual application of featuretools to build an automated project manager, they used random forest feature importances for feature selection. I don’t think it would be too hard to add something like this into the library. Since it’s open-source, the team would probably appreciate the suggestion or any help implementing this!

I’m not sure how well featuretools scales. I know that the company behind featuretools (Feature Labs) uses this at an enterprise level, but I’m not sure of their stack. Maybe you could get in touch, they have been very willing to talk about featuretools!