Informational Technology of Analysis and Forecasting of Number of New Cases of Coronavirus SARS-Cov-2 in Ukraine Based on the Prophet Model

Abstract

The article describes the development of information technology for analysis and forecasting of amount of new confirmed cases of the disease for coronavirus "COVID-19" caused by SARS-CoV-2 infection, based on the daily summary data of the current “wave” in Ukraine, and taking into account various holidays and pseudo-holidays. A review of the known models, which acknowledge such anomalies, was conducted and it is substantiated that considering the current short series of observational data and other conditions, the Facebook Prophet model is optimal for solving this problem. Available data on possible time anomalies in Ukraine in the well-known dataset "COVID-19 Open Data" from Google was characterized, and it is proposed how to take into consideration such anomalies as: public holidays, dates when accordingly to NOAA data weather was warm and without any precipitation, and dates of quarantine easing using information from the "Oxford COVID-19 government response tracker". An algorithm for usage of the proposed information technology was developed, which included a step of two-stage parameter identification and used a separate validation dataset to identify the optimal structure of the model at each stage. Software using Python was created and displayed on Kaggle platform, which then was applied both for Ukraine and for 69 countries around the world. To speed up the research firstly the simplified version of the model was developed with only one stage of parameter identification, and secondly that a dataset "COVID-19: Holidays of countries" was compiled, with information about the holidays of 70 countries, adapted to the needs of this technology and was saved on Kaggle as an open dataset. With the help of the identified models, a number of important conclusions were obtained regarding the understanding of the patterns of coronavirus spread both in Ukraine and in 69 other countries of the world. A model was built to calculate the number of possible new confirmed cases of coronavirus in Ukraine for the next 2 weeks with an error of 2,2 % and using this model, a forecast for the next 2 weeks was made, which was submitted to the Research Group of Mathematical Modeling of Problems Related to the SARS-CoV-2 Epidemic in Ukraine.