Identification of label dependencies for multi-label classification

The main feature distinguishing multi-label classification from a regular classification task is that a number of labels have to be predicted simultaneously. Thus, it is obviously important to exploit potential dependencies between labels. However, surprisingly only a few of the existing algorithms address this problem directly by identifying dependent labels explicitly from the dataset. In this research we propose a new approach for identification and modeling existing dependencies between labels. We define and develop an algorithm that, first, identifies dependencies among the labels, then divides the whole set of labels into several mutually exclusive subsets, and finally performs multilabel classification incorporating the discovered dependencies. In this paper we utilize the ChiSquare test for independence to identify interdependent labels. We then apply a combination of the Binary Relevance and Label Power-set approaches for multi-label classification incorporating the discovered relations. An empirical evaluation on wide range of datasets shows that the proposed method achieves high and stable accuracy results and is competitive with recent state-of-the-art algorithms for multi-label classification