Feature ranking for multi-label classification using predictive clustering trees

In this work, we present a feature ranking method for multilabel data. The method is motivated by the the practically relevant multilabel applications, such as semantic annotation of images and videos, functional genomics, music and text categorization etc. We propose a feature ranking method based on random forests. Considering the success of the feature ranking using random forest in the tasks of classification and regression, we extend this method for multi-label classification. We use predictive clustering trees for multi-label classification as base predictive models for the random forest ensemble. We evaluate the proposed method on benchmark datasets for multi-label classification. The evaluation of the proposed method shows that it produces valid feature rankings and that can be successfully used for performing dimensionality reduction.