A challenge faced while monitoring the COVID-19 pandemic in Brazil is the identification of patterns of incidence and mortality, which can help prioritize interventions to avoid excessive disease transmission and associated deaths. This study aimed to identify epidemiological patterns concerning the evolution of the pandemic among Brazilian federal units (states). The proposed methodology is based on a combination of non-hierarchical k-means clustering and dynamic time warping (DTW), used to measure distances among time series, with the subsequent use of the DTW Barycenter Averaging (DBA) algorithm to calculate cluster centroids for time series of variable lengths. The dataset used is a time series consisting of the number of new cases and deaths per epidemiological week, and the number of cumulative cases and deaths until a given epidemiological week for each of the 27 Brazilian federal units. Six groups of Brazilian federation units were formed based on the similarities between the prevalence and incidence curves. The results demonstrate efficiency with respect to the characterization of both COVID-19 cases and rates of mortality.
Autoria: Jonatas Silva do Espirito Santo, Jackson Santos da Conceição, Lilia Costa, Rosemeire Leovigildo Fiaccone, Marcos Ennes Barreto, Maria Yury Ichihara, Anderson Ara.
Periódico: International Journal of Data Science and Analytics
Ano de publicação: 2024