Two bagging algorithms with coupled learners to encourage diversity

Abstract

In this paper, we present two ensemble learning algorithms which make use of boostrapping and out-of-bag estimation in an attempt to inherit the robustness of bagging to overfitting. As against bagging, with these algorithms learners have visibility on the other learners and cooperate to get diversity, a characteristic that has proved to be an issue of major concern to ensemble models. Experiments are provided using two regression problems obtained from UCI. © Springer-Verlag Berlin Heidelberg 2007.