COMPARING THE ACCURACIES OF FORECASTING MODELS FROM THE TIME SERIES DATA OF COVID-19 INFECTION IN NIGERIA

A variety of forecasting models are now fast becoming among the most important application areas in the analyses of recent COVID-19’s future trends as they provide insight to policy-makers about the development of the disease and on healthcare delivery. However, since there is no one-size-fit-all approach in forecasting the future trends of epidemics, the reliability of these approaches is questioned partly due to time series data characteristics (e.g. quality of the data), uncertainty and nature of the modelling approach (e.g. numerical efficiency of the algorithm). This makes comparison of forecasting models necessary in order to provide an evidence-based information with regards to model performance. This study compared the accuracies of ten models in forecasting the number of population to be affected from Coronavirus in Nigeria (specifically for the whole country, as well as for the Federal Capital Territory (FCT) Abuja and Lagos state). Results show that bagged (bootstrap aggregation) model can provide more consistent accurate results (mean absolute error (MAE) of 48 for Nigeria, 32.80 for Lagos and only 13.48 for FCT) than all models assessed in this study. Other models with good performance include exponential smoothing (Nigeria, MAE = 53.65, Lagos = 36.35, FCT = 14.83), structural time series (Nigeria, MAE = 53.62, Lagos = 34.35, FCT = 14.86), ARIMA (Nigeria, MAE = 53.64, Lagos = 36.34, FCT = 14.83), and theta models (Nigeria, MAE = 53.65, Lagos = 36.35, FCT = 14.83). Although forecasting is challenging as models cannot generally provide accurate daily estimates of the COVID-19 infection, daily COVID-19 cases estimated from these models closely reflect the variation in the original data. The study suggests that combining different approaches is of great value to forecasting modelling and therefore, decision makers should treat results from these approaches with caution and base on analysing scenarios.

Article visualizations: