Visualizing bagged decision trees
J. Sunil Rao, William John Potts · Knowledge Discovery and Data Mining · 1997
We present a visual tablet for exploring the nature of a bagged decision tree (Breiman [1996]). Aggregating classifiers over bootstrap datasets (bagging) can result in greatly improved prediction accuracy. Bagging is motivated as a variance reduction technique, but it is considered a black box with respect to interpretation. Current research seeking to explain why bagging works has focused on different bias/variance decompositions of prediction error. We show that bagging's complexity can be better understood by a simple graphical technique that allows visualizing the bagged decision boundary in low-dimensional situations. We then show that bagging can be heuristically motivated as a method to enhance local adaptivity of the boundary. Some simulated examples are presented to illustrate the technique.