Cost-sensitive feature selection is an important research topic in the field of machine learning and data mining. Presently, cost-sensitive feature selection research is mainly oriented to single-label or multi-label data. Since in many fields of application, there is a correlation and significance among the labels for multi-label data. In order to resolve the problems, this paper introduces label significance into cost-sensitive feature selection, and proposes a feature selection algorithm using test cost based on label significance. The algorithm combines the test cost matrix generated by the three distributions with positive region. Finally, the effectiveness and feasibility of the algorithm are further verified by experimental results on the four Mulan data set.
