A Study on Threshold Selection for Multi-label Classification

Multi-label classification is useful for text categorization, multimedia retrieval, and many other areas. A commonly used multi-label approach is the binary method, which constructs a decision function for each label. For some applications, adjusting thresholds in decision functions of the binary method significantly improves the performance, but few studies have been done on this subject. This article gives a detailed study on the selection of thresholds. Experiments on real-world data sets demonstrate the usefulness of some simple selection strategies.