Browse Prior Art Database

Bias Reduction for the Random Forest Learning Machines

IP.com Disclosure Number: IPCOM000021712D
Publication Date: 2004-Feb-04

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for learning random forest (RF) models with increased predictive accuracy. The disclosed method modifies the split construction algorithm, favoring variables that have better prediction power over the response. Substantial improvement in predictive accuracy is obtained when many irrelevant inputs are present both in classification and regression.