Diversity in Dynamic Class Prediction

Instead of using the same ensemble for all data instances, recent studies have focused on dynamic ensembles in which a new ensemble is chosen from a pool of classifiers for each new data instance. Classifiers agreement in the region where a new data instance resides in has been considered as a major factor in dynamic ensembles. We postulate that the classifiers chosen for a dynamic ensemble should behave similarly in the region in which the new instance resides, but differently outside of this area. In other words, we hypothesize that high local accuracy, combined with high diversity in other regions, is desirable. To verify the validity of this hypothesis we propose two approaches. The first approach focuses on finding the k-nearest data instances to the new instance, which then defines a neighborhood, and maximizes simultaneously local accuracy and distant diversity, based on data instances outside of the neighborhood. The second method makes use of an alternative definition of the neighborhood: all data instances are in the neighborhood. However, the importance of data instances for accuracy and diversity depends on the distance to the new instance. We demonstrate through several experiments that the distance-based diversity and accuracy outperform all benchmark methods.