Comparison of exponential smoothing and ARIMA time series models for forecasting COVID-19 cases: a secondary data analysis

Abstract

Background: In order to manage outbreaks and plan resources, health systems must be capable of accurately projecting COVID-19 case patterns. Health systems can effectively predict future illness patterns by using mathematical and statistical modelling of infectious diseases. Different methods have been used with comparatively good accuracy for various prediction goals in medical sciences. Some illustrations are provided by statistical techniques intended to forecast epidemic cases. In order to increase healthcare systems readiness, this study aimed to identify the most accurate models for COVID-19 with a high global prevalence of positive cases. Methods: Exponential smoothing model and ARIMA were employed on time series datasets to forecast confirmed cases in upcoming months and hence the effectiveness of these predictive models were compared on the basis of performance measures. Results: It was seen that the ARIMA (0,0,2) model is best fitted with smaller values of performance measures (RMSE=4.46 and MAE=2.86) while employed on the recent dataset for short duration. Holt-Winters Exponential smoothing model was found to be more accurate to deal with a longer period of time series based data. Conclusions: The study revealed that working with recent dataset is more accurate to forecast the number of confirmed cases as compared to the data collected for longer period. The early-stage warnings through these predictive models would be beneficial for governments and health professionals to be prepared with the strategies at different levels for public health prevention.