An XGBoost (or any other ensemble model) is a black box, and one can't decipher why the predictions are what they are.
(view the Notebook.md file for the notebook, .Rmd files for the source code)
In this R notebook I demonstrate how to use the inTrees library to interpret and uncover the basic rules influencing the decision trees.
(or atleast my understanding of it)
Flow of inTrees:
Using the Mushroom dataset, classifying mushrooms into poisonous or edible based on various features (Dataset Link)