Cluster analysis of epidemic curves of the third Covid-19 epidemics wave in different regions of Poland

Abstract

Purpose: The purpose of the analysis was to divide Polish regions (as a single region we mean “powiat”) into categories according to the way in which their population passed the third wave of COVID-19 epidemics and the attempt of linking the resultant classifications with other factors. Design/methodology/approach On the basis of data about daily COVID-19 cases per 10,000 inhabitants in regions, weekly averages in every day were calculated. The curves of these averages for each region were approximated by a polynomial function. Then, for the majority of created functions, five characteristic "points" were determined: a maximum of the function together with a maximum and a minimum of its first derivative and two maxima of its second derivative, which are all located the closest to the function maximum. On the basis of coordinates of these points, regions were grouped by using of K-means method. Finally, the mean levels of various factors in obtained categories were analyzed as well as different classification models determined on this basis. Findings: The performed analysis allowed to construct predictive models of the approximate shape of the epidemic curve in a given region. These models can be used in more-depth analysis of epidemics evolution. Research limitations/implications: These models can be used in more-depth analysis of epidemics evolution. Originality/value: By using cross-validation test, the number of clusters equal to five was determined. Mean values of the mentioned coordinates in each cluster allowed to determine an approximate shape of characteristic epidemic curve for a given group of regions. Only among some clusters there was a significant difference in population density, the percentage of population living in cities and the approximate percentage of inhabitants vaccinated after the third wave of COVID-19 epidemics. Nevertheless, on the basis of these factors and the age structure of the population, decision trees which classify most of the wave categories with a satisfactory accuracy were determined. Keywords: Cluster analysis, decision tree, epidemic curve, method of K-means, region. Category of the paper: Research paper.