Evaluation of the Performance of Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Networks (ANNs) Models: Application to Confirmed Cases of Covid-19 in Nigeria

Abstract

In this study, the performance of Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Networks (ANNs) models was investigated and evaluated using daily confirmed cases of COVID-19 in Nigeria. The stationarity status of the data collected was established using Augmented Dickey Fuller unit root test. The residual normality test was also carried out with the residual plots indicating adequacy of the fitted ARIMA model.  The results of neural networks were analyzed using back-propagation for multilayer feed-forward powered by sigmoid function. Utilizing backpropagation method based on three factors expressed in terms of the learning rate, the distance between the actual output and predicted output and the activation function, the network weights were generated  The performance indices for ARIMA and ANNs models were evaluated using Mean Square Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE and the results revealed that the ARIMA model performed better than the ANN model considering the minimum prediction error and forecasting ability. The ARIMA (2, 1, 1)  model  appeared to be the best fitted model over the ANN model for the daily confirmed covid-19 cases considered.