Cooperative coevolutionary instance selection for multilabel problems

Multilabel classification as a data mining task has recently attracted greater research interest. Many current data mining applications address problems having instances that belong to more than one class, which requires the development of new efficient methods. Advantageously using the correlation among different labels can provide better performance than methods that deal with each label separately. Instance-based classification models, such as k-nearest neighbors for multilabel datasets, ML-kNN, are among the best performing methods on any classification task and have also shown very goog performance in multilabel problems. Despite their simplicity, they achieve comparable performance to considerably more complex methods. One of the challenges associated with instance-based classification models is their requirement for storing all the training instances in memory. To ameliorate this problem, instance selection methods have been proposed. However, their application to multilabel problems is problematic because the adaptation of most of their concepts to multilabel problems is difficult. In this paper, we propose a cooperative coevolutionary algorithm for instance selection for multilabel problems. Two different populations evolve together cooperatively. One of the populations is devoted to obtaining solutions for each label, whereas the other population combines these results into solutions for the instance selection for multilabel dataset tasks. On a large set of 70 real-world problems, our approach improves the results of both the ML-kNN method with the whole dataset and an instance selection method using a standard evolutionary algorithm.