WSEAS Transactions on Computers
Print ISSN: 1109-2750, E-ISSN: 2224-2872
Volume 18, 2019
Simplifying Random Forests using Diversity
Authors: ,
Abstract: In this paper, we propose a diversity measure for random forests simplification using both SFS and SBE paths. This is performed in two stages: 1) we use first an overproduce method which generates a large number of trees; 2) We use SFS and SBE paths combined with diversity measurement to reduce the initial ensemble of trees. The proposed method is applied to UCI Repository data sets. A comparative study of the two types of paths with a performance-based pruning method is given. The results are encouraging and allow obtaining ensembles of reduced sizes exceeding, in some cases, the performances of the initial forest and the method used for comparison.