Técnicas e visualização de agrupamentos em aprendizagem não supervisionada com aplicações

Abstract

In the last decade, the amount of information stored in digital format has grown exponentially, leading to the growing need to produce computationally intensive procedures that help generate information from this data. Among others, the area of unsupervised statistical learning provides techniques that help to obtain information from these data without being associated with answers. Thus, this work proposed to study clustering and dimensionality reduction techniques, in order to explain the data configuration based on measuring the distance between objects. In dimensionality reduction, metric and non-metric multidimensional scaling methods were explored to preview the possible groups formed in a reduced dimension space. In turn, for clustering, the K-means, AGNES and DIANA procedures were used. The first, which groups in a partitioned manner, previously requests the number of groups to be formed, while the others, which group hierarchically, bypass this problem in exchange for defining the measure of linkage between groups. Finally, the studied methodologies were applied to real datasets with the R Core Team software (2020).