Using Graph-Based Ensemble Learning to Classify Imbalanced Data

Abstract

The class imbalance problems have attracted considerable attention from researchers of different fields. Ensemble learning has emerged as a powerful approach to address the imbalanced data and improved accuracy and robustness over the single model. In this paper, we present a novel ensemble method based on a bipartite graph (GraphEL) by maximizing the consensus among the multiple binary models. In this bipartite graph, we take into account the probability offered by the multiple classifiers and the average distance provided by the original data, which appear in the graph in the form of weights. Experimental results on 22 imbalanced data sets demonstrate the benefits of the proposed method over the conventional imbalance data handing methods.