The Usefulness of Roughly Balanced Bagging for Complex and High-Dimensional Imbalanced Data
[ 1 ] Instytut Informatyki, Wydział Informatyki, Politechnika Poznańska | [ P ] employee
2016
paper
english
- class imbalance
- roughly balanced bagging
- types of minority examples
- high-dimensional data
- random subspace method
EN Under-sampling generalizations of bagging ensembles improve classification of imbalanced data better than other ensembles. Roughly Balanced Bagging is the most accurate among them. In this paper, we experimentally study its properties that may influence its good performance. Results of experiments show that it can be constructed with a small number of component classifiers. However, they are less diversified than components of the standard bagging. Moreover, its good performance comes from its ability to recognize unsafe types of minority examples better than other ensembles. We also present how to improve its performance by integrating bootstrap sampling with the random selection of attributes.
93 - 107
WoS (15)