The aim of this study is to contribute to the literature by estimating the 5-weeks number of cases/deaths for each continent by using statistical-based prediction models, which are quite effective on simple but small-scale datasets. While Auto.arima, Tbats, Naive, Holt, Thetaf and, Drift models were used for prediction processes root mean square error (RMSE), mean absolute error (MAE), and mean absolute percent error (MAPE) metrics were used for evaluating estimates. According to the confirmed cases MAPE metric values of the 5 continents analyzed, the best predictions for Asia, Africa, Europe, America, and Oceania were done by Thetaf, Naive, Thetaf, Auto.arima, and Auto.arima models, respectively. The use of very limited data for time series estimates such as 57-weeks in the estimation process was a disadvantage. Most models require at least two cycles, 104 weeks of data, to run. Therefore, we could not use models such as neural network autoregressive, multilayer perceptrons, extreme learning machines. The results obtained with the prediction models used in this study aim to make more accurate decisions for the authorized persons dealing with health to be more prepared for future conditions and health systems.

Reviewers Suggestions


zpala_DUJE_final.docx (2021 kB)
Manuscript Final

ResponseToReviewer1_Rev.pdf (85 kB)
Response to Reviewer 1

ResponseToReviewer2_Rev.pdf (520 kB)
Response to Reviewer 1