Mineração de dados temporais da covid-19 no Brasil: busca de similaridades nas características dos estados brasileiros através do agrupamento hierárquico

Abstract

The COVID-19 pandemic posed a serie of unprecedented challenges to global public health, with Brazil being one of the severely affected countries. Various studies have been conducted to compare different temporal contexts of COVID-19. This work carries out a similar exploration, delving into the temporal dynamics of COVID-19 in Brazil, and seeking potential patterns among specific states by comparing them based on certain characteristics. For this purpose, a dataset containing notifications of disease cases in each Brazilian state during the years 2020 and 2021 was used, and distance metrics such as Dynamic Time Warping (DTW) and Euclidean Distance were employed to analyze time series data. Furthermore, tools like PySpark and Pandas were utilized for data manipulation and preparation, thus facilitating subsequent analysis. Methods such as feature importance were applied to identify and select characteristics in the notifications that impacted clinical evolutions, and Hierarchical Clustering was used to relate the time series based on their similarities.