An evaluation of classifier ensembles for class imbalance problems

Classification of imbalanced data represents a challenging task in machine learning, as most classification algorithms tend to bias towards the majority class, while often correctly identifying minority class instances is of greater importance. Consequently, there is a need for methods that provide improved accuracy for the minority class without sacrificing overall performance. Ensemble classification methods have been shown to be able to lead to both more robust as well as better performing classification approaches, while more recently, ensemble approaches have also been developed for imbalanced classification problems. In this paper, we evaluate seven state-of-the-art ensemble approaches and, in an extensive set of experiments, compare their performance on five benchmark datasets. Our results should allow to shed some light on strengths and weaknesses of the investigated algorithms.