A transformation approach towards big data multilabel decision trees

Abstract

A large amount of the data processed nowadays is multilabel in nature. This means that every pattern usually belongs to several categories at once. Multilabel data are abundant, and most multilabel datasets are quite large. This causes that many multilabel classification methods struggle with their processing. Tackling this task by means of big data methods seems a logical choice. However, this approach has been scarcely explored by now. The present work introduces several big data multilabel classifiers, all of them based on decision trees. After detailing how they have been designed, their predictive performance, as well as the execution time, are analyzed.