Regressão multi alvo via agrupamento hierárquico das variáveis dependentes

Abstract

Multi-target regression is a crucial technique in machine learning, applied to problems where multiple dependent variables need to be predicted simultaneously. In this work, a new approach is proposed that utilizes hierarchical clustering of the dependent variables to explore and model the complex relationships between the multiple targets. The methodology involves identifying underlying hierarchical structures in the output data, enabling more accurate and interpretable modeling of the interdependencies among the dependent variables. The approach was evaluated through extensive experiments on various datasets, comparing its performance with traditional multi-target regression methods. The main contribution is the introduction of hierarchical clustering techniques in the context of multi-target regression, providing a versatile framework that can be applied across different domains. Despite promising results, statistical analyses indicate that the use of hierarchical clustering did not show statistically significant variation in the efficacy of multi-target regression, highlighting the need for the use of other dissimilarity metrics.