Comparative Predictive Performance of Artificial Neural Networks and ARIMA Models for COVID-19 Case Forecasting in Nigeria
Hakeem Temitope Adekunle
*
Department of Statistics, Federal University of Technology, Akure, P.M.B. 704, Akure, Ondo State, Nigeria.
Olusoga Akin Fasoranbaku
Department of Statistics, Federal University of Technology, Akure, P.M.B. 704, Akure, Ondo State, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
COVID-19, a novel strain of coronavirus first identified in December 2019 in Wuhan, China, quickly escalated into a global pandemic. In response, various predictive models have been employed to monitor and forecast its spread. This study compares the predictive performance of two such models Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Network (ANN) using COVID-19 daily case data in Nigeria from March 23, 2020, to April 23, 2021, sourced from the Nigeria Centre for Disease Control (NCDC) and Our World in Data. The dataset was partitioned into 80:20 and 60:40 training-testing ratios. Results showed that ANN significantly outperformed ARIMA in both scenarios. For 80% training and 20% testing, ANN achieved RMSE = 0.00164 and MAE = 0.00024, while ARIMA recorded RMSE = 0.85758 and MAE = 98.52. Similarly, with 60% training and 40% testing, ANN achieved RMSE = 0.00049 and MAE = 0.00157, compared to ARIMA's RMSE = 0.82258 and MAE = 57.29. The findings indicated that for prediction purposes, neural networks should be considered, and for efficiency with large samples and significant training data, neural networks should also be taken into account.
Keywords: COVID-19, ARIMA, artificial neural network, forecasting, predictive modeling