The data mining and machine learning research communities have focused on developing specialized
algorithms and methods to handle a multitude of potential complications which may confront the traditional supervised learning task. Of these, class imbalance is among the most persistent in real-world applications. Less attention has been garnered for the set of problems in which the data distribution changes, potentially wiping out the gains from expensive data mining methods. To an even lesser degree has the combination of problems been considered. It is the purpose of this dissertation to explore concepts of distributional
change, particularly within the context of imbalanced data problems and the effects of the performance on solutions from this realm. Based on this exploration, the proposed dissertation will derive methods to identify and handle both problems simultaneously.
