Correlation analysis of performance measures for multi-label classification
作者：
Highlights：
• Evaluation measures have been used arbitrarily in multilabel classification experiments, without an objective analysis of correlation or bias.
• A comprehensive and detailed analysis of the correlation that exists between multilabel measures is presented.
• Hamming Loss is a highly recommended measure, as it is not correlated with others and it is the most employed measure in the literature.
• At least 12 out of 16 multilabel measures adopted in the literature are highly correlated with each other.
摘要
•Evaluation measures have been used arbitrarily in multilabel classification experiments, without an objective analysis of correlation or bias.•A comprehensive and detailed analysis of the correlation that exists between multilabel measures is presented.•Hamming Loss is a highly recommended measure, as it is not correlated with others and it is the most employed measure in the literature.•At least 12 out of 16 multilabel measures adopted in the literature are highly correlated with each other.
论文关键词：Multi-label classification,Evaluation measures
论文评审过程：Received 23 September 2014, Revised 19 August 2017, Accepted 4 January 2018, Available online 16 January 2018, Version of Record 16 January 2018.
论文官网地址：https://doi.org/10.1016/j.ipm.2018.01.002
