Abstract: Multilabel classification is a type of supervised learning that has gained significant interest in recent years. Multilabel classification addresses scenarios where each sample receives multiple binary classifications, known as labels in this context. Learning multilabel datasets is more challenging than single-label classification. In this paper, we propose a new problem transformation method that is based on considering from complex relationships among the labels at the top level to very simple ones at the bottom level. We propose a hierarchical model that partitions the set of labels into increasingly smaller clusters. Our classifier is a combination of all the classifiers constructed from each labelset obtained by voting. That way, our proposal benefits from the high-order approaches of the top level compared to the low-order approaches of the bottom levels. A comparison against 10 top-performing multilabel methods using 14 performance metrics and 60 datasets shows the advantage of our approach.
N. García-Pedrajas, and G. Cerruela-García (2024) “PYRAMID: A label hierarchical clustering approach for multi-label classification,” submitted.
Detailed results and additional tables