I. Introduction
Multilabel classification has recently received widespread attentions due to its increasing applications, such as in text categorization [1], [2], image classification [3]–[5], bioinformatics [6], [7], and medical diagnosis [8], [9]. In traditional single-label classification tasks, each data object is associated with a single label. In contrast, in multilabel classification tasks, each data object is associated with a subset of labels, called “relevant labels,” while the remaining labels are called “irrelevant labels.” Often, there exists a correlation among labels. For example, an image with the labels “sun” and “beach” is more likely related to the label “ocean” but less likely related to the label “rain.” The objective of multilabel classification is to recognize relevant labels from a predefined label set for a given input.