Which of the statements below regarding bagging/random forest approaches is the most accurate?

Group of answer choices

A) In these approaches we first fit a deep (likely overfitting) tree to the data, as tightly as we can, then introduce additional term into the score function that penalizes the number of leaves in the tree (i.e. the number of the resulting regions); next, we minimize such modified score function by trying to remove different splits from the tree. A collection of such trees obtained for different penalties is a random forest, and we make the prediction by takingthe predictions of each such differently penalized model and averaging them.

B) In these approaches we first fit a deep (likely overfitting) tree to the data, as tightly as we can, and then repeatedlybootstrap that model, by randomly selecting different combinations and orders of the splits (i.e. the branching nodes) from that tree. We hope that by trying different splits in different orders, with replacement, and averaging predictions of the resulting models, we can obtain a better overall prediction

C) In these approaches we repeatedly search for the best possible fit to different bootstrapped datasets. As the result, different aspects and features of the data are randomly in or out of, or more or less prominent in, any particular bootstrapped dataset Di used to train model Mi. This mimics, to a certain extent, drawing a new sample every time from the true underlying distribution (assuming the original sample we have at hand is representative of course!), so by repeating this procedure and averaging the predictions of models Mi we expect obtaining a better prediction.

D) In these approaches we repeatedly search for different somewhat imperfect fits to the data we have at hand. We hope that while each model Mi is deliberately imperfect, they might have smaller variance and by averaging the predictions of all these models we can get a better prediction at the end.