Development of Algorithms for Choosing the Best Time Series Models and Neural Networks to Predict COVID-19 Cases

Time series analysis became one of the most investigated fields knowledge during spreading COVID-19 around world. The problem modeling and forecasting infection cases COVID-19, deaths, recoveries other parameters is still urgent. Purpose study. Our article devoted to investigation classical statistical neural network models that can be used for cases. Materials methods. We discuss model NNAR, compare it with linear nonlinear (BATS, TBATS, Holt's trend, ARIMA, epidemiological SIR model). In our we Epemedic.Network algorithm using R programming language. This takes time as input data chooses best from SIR, model. selection criterion MAPE error. consider implementation COVID -19 in Chelyabinsk region, predicting possible peak third wave three scenarios. mention considered work any se-ries, not only ones. Results. developed helped identify pat-tern region realized parts consi-dered algorithm. It should noted make form short-term forecasts sufficient accuracy. show increase number neurons led in-creasing accuracy, there are where error reduced case reducing neurons, this depends on pattern. Conclusion. Hence, get a very accurate forecast, recommend re-running weekly. For medium-range fore-casting, NNAR among those but also allows good horizon 1–2 weeks.