Forest CERN: A New Decision Forest Building Technique

Persistent efforts are going on to propose more accurate decision forest building techniques. In this paper, we propose a new decision forest building technique called “Forest by Continuously Excluding Root Node (Forest CERN)”. The key feature of the proposed technique is that it strives to exclude attributes that participated in the root nodes of previous trees by imposing penalties on them to obstruct them appear in some subsequent trees. Penalties are gradually lifted in such a manner that those attributes can reappear after a while. Other than that, our technique uses bootstrap samples to generate predefined number of trees. The target of the proposed algorithm is to maximize tree diversity without impeding individual tree accuracy. We present an elaborate experimental results involving fifteen widely used data sets from the UCI Machine Learning Repository. The experimental results indicate the effectiveness of the proposed technique in most of the cases.