Regularization Methods in Random Forests

Abstract

Shrinkage is still a hard task to perform in tree-based methods. In this work, we evaluated and extended the current methods. Our extension has demonstrated to be better than what was proposed in (Deng and Runger, 2012), since we ended up with fewer variables but still a good performance. Nevertheless, we see that the method still has a lot of room for improvement, which defines our next steps.

Date
Location
Maynooth, Ireland