I built an XGB and ran RFECV over 250 features. After an hour or so, I plotted the grid_scores_.
All numbers of features are within 0.02, as clearly visible on the y-axis.
To me this plot would indicate that it doesn't matter if I have 1 variable or 250 in my model. Which has a lot of scary ramifications.
Has anyone ever come across this in their work? If so, what did you determine? How is this even possible?
