Acessibilidade / Reportar erro

Stability in cluster analysis: study of case in forest science

The main objective of this research was to propose a system to the study and interpretation of stability in cluster analysis through several cluster algorithms in vegetation data. The data set used derived from a survey in the Silviculture Forest at Federal University of Viçosa - MG. To perform the cluster analysis, the Mahalanobis distance matrices were estimated on basis of original data and bootstrap resampling. Also, the single linkage, complete linkage, average distances, centroid, median and Ward methods were used. Chi-square test was applied to detect the association among the methods. A co-phenetic correlation was obtained for the cluster methods. The results for the method associations were very similar, indicating that any algorithm of the studied clusters is stabilized, and in fact, that groups exist among the analyzed individuals. However, it was verified that the methods are coincident, except for the centroid and Ward, and also the centroid and median methods, when compared to Ward, respectively, based on the Mahalanobis matrices derived from the original data set and bootstrap. The methodology proposed is promising to the study and interpretation of the stability of cluster analysis methods in vegetation data.

Multivariate analysis; bootstrap; agglomerative hierarchical methods


Sociedade de Investigações Florestais Universidade Federal de Viçosa, CEP: 36570-900 - Viçosa - Minas Gerais - Brazil, Tel: (55 31) 3612-3959 - Viçosa - MG - Brazil
E-mail: rarvore@sif.org.br