Análise de agrupamento hierárquicos

Abstract

Hierarchical cluster analysis is an effective way to group elements together to facilitate exploratory data analysis. The groups are selected so that within each group the elements are homogeneous and the groups are heterogeneous to each other. Its importance lies in the need to establish a profile and thus understand the behavior of each group. First, the similarity and association coefficients were presented, used to calculate distances with different types of variables. In view of this, the hierarchical methods were differentiated, which are: single link method, complete link, centroid, average of distances and ward, with their particularities. The Agnes and Diana algorithms were also presented. Then, methods of adjustment and choice of the ideal number of groups were presented. Finally, the R software was used to demonstrate the methods and compare the euclidean and manhattan distances.