Final thoughts

• Bagging improves the prediction accuracy for high variance (and low bias) models

• VIPs and PDPs can help to make inferences about the how model leverages feature information.

• It’s easy to do in parallel as it performs independent processes.

But

• The trees are not completely independent of each other since all the original features are considered at every split of every tree, returning correlated results which stop the model for further reducing the variance.