Análise do índice de nebulosidade para otimização do processo de agrupamentos de dados

Abstract

The technique of clustering analysis is an important tool in scientific research, it can be used in various fields of knowledge such as medicine, biology and statistics. To group data in clusters is a way to reflect the internal data structure and identify classes present in this clusters so within the same class there is homogeneity and there is heterogeneity between different classes. There are three types of clustering methods used to find optimal partitioning: hierarchical methods, methods based on graph theory and methods based on objective function. In this study we used the objective function algorithm based on Fuzzy C-Means and also the bootstrap resampling technique. The idea is to vary the cloudiness index in order to find the best value to be used for sorting the databases: Iris, Wine and three other artificial databases, consequently obtaining better partitioning results. The quality of the partitioning is based on traditional measures of comparison such as Crusade Classification (Acc), F1, Hubert (Hub), Jaccard, Random index (Rand) and Fowlkes and Mallows (Fowlkes). The results obtained so far show that the best range for the cloudiness index is between 1.04 and 1.2 for the contents of measures adopted.