COMPARISON OF CROUPING TECHNIQUES: A CASE STUDY ON VACCINATION AND CHILD MORTALITY DATA BETWEEN 2011 AND 2021

Abstract

The present work aims to show how cluster analysis can be carried out, using the hierarchical technique and not hierarchy. Data on vaccination and infant mortality of children aged between 28 days and one year of life were used in a time frame ranging from 2011 to 2021, available in DATASUS of the Brazilian states through the numbers of infected people in each state, in order to identify similarities between states through the numbers of infected people, offering a counterpoint to the criterion used to analyze the number of infected people in states, based on the size of the population and comparing it with their population. For cluster analysis, the Euclidean matrix was used with the hierarchical method, the simple, complete, average, ward linkage methods and a non-hierarchical method through the K-means method were applied, the methods of determining the ideal number of groups using methods such as elbow, average silhouette coefficient and Rand adjusted conphenetic correlation coefficient to measure the degree of adjustment between the original matrices and the matrix resulting from the simplification provided by the dendrogram. However, the method that best represents the data was found to be Ward's. When grouping the states of both data, the similarity between the data variables and the correlation were taken into account, where it can be seen that the data are correlated.