| Abstrakt: |
In this work, mathematical models of the spread of the coronavirus COVID-19 in various countries are built, and a comparative analysis of these models for the United States, Mexico, Russia, Belgium and Ukraine was performed. Baseline data on the number of infections obtained from the daily reports of the World Health Organization and the the Center for Systems Science and Engineering at Johns Hopkins University. To simulate the spread of coronavirus, two powerful classes of machine learning methods have been selected that allow predicting nonlinear time series: support vector machines and feed forward multilayer neural networks. The advantages and disadvantages of these methods are revealed, and the issues of regularization are considered. The construction and training of time series models to describe the spread of COVID-19 in different countries, the choice of the best model, the construction of forecast and the visualization of results were performed in an implemented software module in the python environment using modern scikit-learn, pandas and matplotlib libraries. Using the grid search method with cross-validation, the best parameters of neural network and support vector models which describe the spread of COVID-19 in the USA, Mexico, Russia, Belgium and Ukraine were selected. Based on the constructed models, the growth of COVID-19 diseases in these countries was predicted. [ABSTRACT FROM AUTHOR] |
