PROCEDURE FOR THE FORMATION OF GROUPS OF COMPANIES AND FOR BUILDING THE INDEX FOR CLUSTER ASSESSMENT

The investigation of similarity patterns between companies can be accomplished by the formation of similarity groups. However, there is always doubt about the best way to build clusters. Thus, the objective of this work was to study some clustering methods and some possible group sizes, so as to obt...

Descripción completa

Detalles Bibliográficos
Autores: Seidel, Enio Júnior, Oliveira, Marcelo Silva de, Tavares, Bruno, Antonialli, Luis Marcelo
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2012
País:Brasil
Institución:Universidade Federal Fluminense (UFF)
Repositorio:Sistemas & Gestão
Idioma:portugués
OAI Identifier:oai:ojs.www.revistasg.uff.br:article/151
Acceso en línea:https://www.revistasg.uff.br/sg/article/view/V7N1A6
Access Level:acceso abierto
Palabra clave:Cluster Analysis
Discriminant Analysis
Group Sizes
Index for Evaluating Clusters
Análise de agrupamentos
Análise discriminante
Tamanhos de grupos
&Iacute
ndice de avaliação dos agrupamentos.
Descripción
Sumario:The investigation of similarity patterns between companies can be accomplished by the formation of similarity groups. However, there is always doubt about the best way to build clusters. Thus, the objective of this work was to study some clustering methods and some possible group sizes, so as to obtain a good clustering. In addition, an index for evaluating the clusters formed was proposed. Six clustering methods and three group sizes were considered. After the application of cluster analysis, the groups formed were assessed by the discriminant analysis, the Kappa coefficient and an index developed to assess the initial clusters. Overall, it was found that the best option would be to apply the MW method with three groups and after, to apply the discriminant analysis to obtain an appropriate final number of companies per group. This procedure allowed obtaining groups relatively similar in terms of number of elements per group. Furthermore, an interesting alternative is to make use of the index for assessing initial clusters to select the clustering method to be used, allowing selecting the best procedure.