I'm doing some stacked generalization/meta-learning.
In blogs and posts, I have only seen people take the level 1 predictions and just directly use them as features for a level 2 model (no feature engineering). However, I recognize that there is a possibility for feature engineering here, as you could (for example) take the average of the level 1 predictions, take the standard deviation of them (their level of agreement/disagreement with one another) and potentially other transformations/combinations.
I'm wondering if there are any common types of features that are engineered from the predictive outputs of the level 1 models?
