Uma estratégia híbrida para o problema de classificação multirrótulo

Abstract

This work presents a new method for multi-label classification based on Particle Swarm Optimization, called Multi Label K-Nearest Michigan Particle Swarm Optimization ML-KMPSO) and evaluates it experimentally using two real-world datasets. Multilabel learning first arose in the context of text categorization, where each document may belong to several classes simultaneously. In this work, we propose a new hybridapproach, ML-KMPSO. It is based on two strategies. The first strategy is the Michigan Particle Swarm Optimization (MPSO), which breaks the multi-label classification task into several binary classification problems, but it does not take into account the correlations among the various classes. The second strategy is Multi Label K-Nearest Neighbor (ML-KNN), which is complementary and takes into account the correlations among classes. We evaluated the performance of ML-KMPSO using two real-world datasets: Yeast gene functional analysis and natural scene classification. The experimental results show that ML-KMPSO produced results that match or outperform well-established multi-label learning algorithms.