Cluster-Based Instance Selection for the Imbalanced Data Classification

Instance selection, often referred to as data reduction, aims at deciding which instances from the training set should be retained for further use during the learning process. Instance selection is the important preprocessing step for many machine leaning tools, especially when the huge data sets are considered. Class imbalance arises, when the number of examples belonging to one class is much greater than the number of examples belonging to another. The paper proposes a cluster-based instance selection approach for the imbalanced data classification. The proposed approach bases on the similarity coefficient between training data instances, calculated for each considered data class independently. Similar instances are grouped into clusters. Next, the instance selection is carried out. The process of instance selection is controlled and carried-out by the team of agents. The proposed approach is validated experimentally. Advantages and main features of the approach are discussed considering results of the computational experiment.