A Semi-Random Subspace Method for Classification Ensembles

Abstract

The performance of ensemble algorithms is related with two terms: the individual accuracy of base learners and the diversity of their results. Random Subspace algorithm owes its success to the diversity. In this study, we propose a method (Semi Random Subspace) which increases its diversity. We compare our method and original Random Subspace over 36 datasets. The experiments show that our method is superior to the original Random Subspace. But its advantage is limited with the size of the ensemble. In this situation, we can say that Semi Random Subspace is suitable choice for the small ensembles.