Importance of careful data tracking for short-term prediction of COVID-19 case and death numbers from official resources

Abstract

During the COVID-19 outbreak, governments, scientists, health workers, and numerous people worked on strategies or solutions for halting disease propagation. Unfortunately, the need for monitoring is steeply increasing, and taking necessary and restrictive actions is currently unavoidable. Due to the lack of epidemiological data and constantly changing numbers, constructing less error-prone predictive models and reliable mathematical models for the near future will help make better legal actions and prevention strategies. 
 In this study, daily data of eleven countries between 01/21/2020-05/02/2020 and 01/21/2020-06/17/2020 were used to forecast the number of future COVID-19 events by using different forecasting models. Best fit models were chosen after analysis of present numbers with Auto-Regressive Integrated Moving Average(ARIMA), Brown’s LES, and Holt’s LES models based on MAPE values. 
 The study showed the least error-prone best-fit models for short-term future predictions by analyzing two datasets and demonstrated that models changed after data updates among the selected countries. Investigation of the data of USA (Holt’s MAPE=7,7 to ARIMA(2,2,0) MAPE=4,8 for case numbers and ARIMA (2,0,0) MAPE=5,7 to ARIMA(1,2,0) MAPE=3,4 for death numbers), Turkey (ARIMA(2,0,0) MAPE=4,0 to Brown's LES MAPE=1,7 for case numbers and ARIMA(2,1,1) MAPE=0,9 to ARIMA(0,2,0) MAPE=0,9 death numbers), Brazil (Holt's MAPE=6,2 to ARIMA(1,0,1) MAPE=36,4 for case numbers and Brown's MAPE=3,2 to ARIMA(1,2,0) MAPE=2,8 for death numbers), Russia (ARIMA(1,2,0) MAPE=6,8 to ARIMA(1,2,0) MAPE=3,5 for case numbers and ARIMA(1,1,1) MAPE=3,7 to ARIMA(2,2,0) MAPE=3,5 for death numbers) demonstrated that at the same time flow, updating data caused alterations in the model selection, which results with changes in the predictions.
 The results of this study indicate that using more than one statistical model has superiority in the current approaches and fluctuations in the numbers should be taken into account when using the data to construct mathematical models and create future predictions for the management of already complicated and exhausting COVID-19 pandemic. Thus, policies and restrictions against COVID-19 spread might be more successful after considering that adjusted predictions for providing more accurate results.