Abstract: Rakel (Random k-labelsets) randomly selects a number of label subsets from the original set of labels and uses the LP (Label Powerset) method to train the corresponding multi-label classifiers. But the models maybe have a poor performance because of randomization nature. Thus in this paper we firstly capture some pairwise relationships based on label co-occurrence between the labels to training LP classifier by PwRakel (Pairwise Random k-labelsets) algorithm. The method extends the training set by exploiting label correlations to improve classification performance effectively. The experimental results indicate that the proposed method improves multi-label classification accuracy compared with the Rakel algorithm and to other state-of-the-art algorithms.

Key words: multilabel classification, label correlation, PwRakel