Portal de Periódicos da CAPES

Artigo Produção Nacional Revisado por pares

BIBTEX RIS

A generalized average linkage criterion for Hierarchical Agglomerative Clustering

2020; Elsevier BV; Volume: 100; Linguagem: Inglês

10.1016/j.asoc.2020.106990

1872-9681

Leonardo Ramos Emmendörfer, Anne M. P. Canuto,

Data Mining Algorithms and Applications

Hierarchical agglomerative clustering (HAC) is among the most widely adopted algorithms in unsupervised learning. This method employs a linkage criterion to measure the similarity between two clusters and the correct selection of this criterion highly influences the performance of HAC. This paper presents and evaluates a novel linkage criterion for HAC, which is a generalization of the Average Linkage (GAL) and it aims at improving the quality of the similarity computation of the original average linkage criterion. In order to assess the liability of the proposed criterion, an empirical analysis is conducted, which is performed on 28 datasets from the literature. In a comparative analysis, the proposed criterion is compared to seven reference methods from the literature. Our findings indicate that the results obtained by the proposed criterion are promising, surpassing all the existing reference methods.