Since instances in multi-label problems are associated with several labels simultaneously, most traditional feature selection algorithms for single label problems are inapplicable. Therefore, new criteria to evaluate features and new methods to model label correlations are needed. In this paper, we adopt the graph model to capture the label correlation, and propose a feature selection algorithm for multilabel problems according to the graph combining with the large margin theory. The proposed multi-label feature selection algorithm GMBA can efficiently utilize the high order label correlation. Experiments on real world data sets demonstrate the effectiveness of the proposed method. The codes of the experiment of this paper are available at https://github. com/Faustus-/ECML2016-GMBA.
CITATION STYLE
Yan, P., & Li, Y. (2016). Graph-margin based multi-label feature selection. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9851 LNAI, pp. 540–555). Springer Verlag. https://doi.org/10.1007/978-3-319-46128-1_34
Mendeley helps you to discover research relevant for your work.