B1813
Title: Statistical inferences for measures of multi-label classification
Authors: Kanae Takahashi - Hyogo Medical University (Japan) [presenting]
Abstract: Data classification problems can be categorized into single-label classification and multi-label classification. In single-label classification, the data are mutually exclusive and are classified into exactly one of the classes. In multi-label classification, on the other hand, data are not mutually exclusive and can be classified into several classes simultaneously. Several evaluation measures have been proposed for single-label and multi-label classifications. While the interval estimation and hypothesis testing method have been proposed for evaluation measures of single-label classification, point estimation can only be performed for evaluation measures of multiple-label classification, and no interval estimation method has yet been proposed. To address these knowledge gaps, statistical inference methods are proposed for evaluation measures of multi-label classifications. The performance of the proposed methods is investigated through simulations.
