Abstract: Multi-label classification is a growing paradigm in the fields of data mining and machine learning. In multi-label learning, each sample can belong to more than one binary class, termed label in the multi-label framework, in contrast with standard single label learning. The challenge of producing a better classifier lies in the advantageous use of the correlation among the different labels. In recent years, many multi-label models have been proposed that make the decision of which methods to use troublesome. In this paper, we present the most comprehensive comparison carried out thus far centered around the study of ranking performance. We conduct a comprehensive analysis by comparing 56 distinct methods and several configurations of each method, resulting in a total of 173 trained models. In addition, we utilize an extensive collection of problems consisting of 65 datasets. Furthermore, we analyze the effectiveness of the techniques by evaluating their performance using six different ranking performance metrics. Our findings indicate that while certain strategies consistently rank well among the top-performing models, the most effective methods are strongly correlated with the specific metric used to assess performance. Furthermore, we examine many aspects of the approaches behavior.
N. García-Pedrajas, José M. Cuevas-Muñoz, G. Cerruela-García and A. de Haro-García (2023) “Multi-label methods extensive experimental comparison focused on ranking performance,” submitted.
Detailed results and Figures