摘要：A investigação de padrões de semelhança entre empresas pode ser realizada através da construção de grupos por similaridade. Contudo, geralmente se tem dúvidas quanto à melhor forma de construir os agrupamentos. Desse modo, tem-se o objetivo de estudar alguns métodos de agrupamentos e alguns possíveis tamanhos de grupos, de maneira a se obter um bom agrupamento. Além disso, se propõe um índice de avaliação dos agrupamentos formados. Consideraram-se seis métodos de agrupamentos e três tamanhos de grupos. Após a aplicação da análise de agrupamentos, foi realizada a avaliação dos agrupamentos formados através da análise discriminante, do coeficiente de concordância Kappa e de um índice desenvolvido para avaliar os agrupamentos iniciais. De forma geral, verificou-se que a melhor possibilidade seria aplicar o método MW com três grupos e, após, aplicar a análise discriminante até obter um número final adequado de empresas por grupo. Com este procedimento se obtêm grupos relativamente semelhantes em termos de número de elementos por grupo. Além disso, uma alternativa interessante é tomar uso do índice de avaliação dos agrupamentos iniciais para a escolha do método de agrupamento a ser utilizado, permitindo fazer a opção pelo melhor procedimento.↓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.
其他摘要：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.
关键词：Sistemas produtivos; Gestão; Estatística aplicada;Análise de agrupamentos; Análise discriminante; Tamanhos de grupos; Índice de avaliação dos agrupamentos.;Cluster Analysis; Discriminant Analysis; Group Sizes; Index for Evaluating Clusters
其他关键词：Cluster Analysis; Discriminant Analysis; Group Sizes; Index for Evaluating Clusters
