Unsupervised analysis of COVID-19 pandemic evolution in brazilian states

Extracting information and discovering patterns from a massive dataset is a hard task. In an epidemic scenario, this data has to be integrated providing organization, agility, transparency and, above all, it has to be free of any type of censorship or bias. The aim of this paper is to analyze how coronavirus contamination has evolved in Brazil applying unsupervised analysis algorithms to extract information and find characteristics between them. To achieve this goal we describe an implementation that uses data about Covid-19 spread in Brazilian states (26 states and the federal district), applying a Time Series Clustering technique based on a K-Means variation, using Dynamic Time Warping as a similarity metric. We used data reported by the Ministry of Health in Brazil, referring to deaths per 100k inhabitants, during 452 days from the first reported death in each state. Two analyzes were performed, one considering 3 clusters and the other with 6 clusters. Through these analysis, 3 patterns of responses to the pandemic can be observed, ranging from one of greater to lesser control of the pandemic, although in recent months all clusters showed a highly increase in the number of deaths. The identification of these patterns is important to highlight possible actions and events, as well as other characteristics that determine the correct or incorrect public decision-making in combating the Covid-19 pandemic.