Data reduction and stacking for imbalanced data classification

Abstract

Class imbalance arises when the number of examples belonging to one class is much greater than the number of examples belonging to another. The discussed approach focuses on combining several techniques including data reduction and stacking with the aim of improving the performance of the machine classification in the case of imbalanced data. The paper proposes a cluster-based data reduction approach assuming that the instances are selected from a cluster, the data reduction is carried-out on instances belonging to the majority classes, and the aim of the instance selection is to reduce the imbalance ratio between the majority and minority classes. The process of instance selection is carried out with using an agent-based population learning algorithm. To increase performance and generalization ability of the prototype-based machine learning classification it was decided to use the stacking technique. The proposed approach is validated experimentally using several benchmark datasets from the KEEL repository. Advantages and main features of the approach are discussed considering the results of the computational experiment.