Application of artificial neural networks to predict the COVID-19 outbreak

Hamid Reza Niazkar,Majid Niazkar

doi:10.1186/s41256-020-00175-y

Hamid Reza Niazkar, Majid Niazkar

Open Access

https://doi.org/10.1186/s41256-020-00175-y

Copy DOI

Abstract

BackgroundMillions of people have been infected worldwide in the COVID-19 pandemic. In this study, we aim to propose fourteen prediction models based on artificial neural networks (ANN) to predict the COVID-19 outbreak for policy makers.MethodsThe ANN-based models were utilized to estimate the confirmed cases of COVID-19 in China, Japan, Singapore, Iran, Italy, South Africa and United States of America. These models exploit historical records of confirmed cases, while their main difference is the number of days that they assume to have impact on the estimation process. The COVID-19 data were divided into a train part and a test part. The former was used to train the ANN models, while the latter was utilized to compare the purposes. The data analysis shows not only significant fluctuations in the daily confirmed cases but also different ranges of total confirmed cases observed in the time interval considered.ResultsBased on the obtained results, the ANN-based model that takes into account the previous 14 days outperforms the other ones. This comparison reveals the importance of considering the maximum incubation period in predicting the COVID-19 outbreak. Comparing the ranges of determination coefficients indicates that the estimated results for Italy are the best one. Moreover, the predicted results for Iran achieved the ranges of [0.09, 0.15] and [0.21, 0.36] for the mean absolute relative errors and normalized root mean square errors, respectively, which were the best ranges obtained for these criteria among different countries.ConclusionBased on the achieved results, the ANN-based model that takes into account the previous fourteen days for prediction is suggested to predict daily confirmed cases, particularly in countries that have experienced the first peak of the COVID-19 outbreak. This study has not only proved the applicability of ANN-based model for prediction of the COVID-19 outbreak, but also showed that considering incubation period of SARS-COV-2 in prediction models may generate more accurate estimations.

Highlights

The novel coronavirus (COVID-19 or SARS-COV-2) epidemic has involved into a global pandemic
Each artificial neural networks (ANN)-based model is first ranked in accordance with their values obtained for Mean Absolute Relative Error (MARE), Normalized Root Mean Square Error (NRMSE) and R2
Such database is not available for the COVID-19 outbreak in this time interval. These shortcomings may confine the use of ANN for predicting the COVID-19 outbreak, while this study proposed fourteen novel ANN-based models for this purpose

Summary

Introduction

The novel coronavirus (COVID-19 or SARS-COV-2) epidemic has involved into a global pandemic. Niazkar and Niazkar [13] exploited multi-gen genetic programming, which is one of AI models, to develop mathematical models with the exponential function for predicting the COVID-19 pandemic in seven countries including China, Republic of Korea, Japan, Italy, Singapore, Iran and United States of America. Li et al [14] suggested an exponential function to predict the trend of the COVID-19 outbreak They estimated the end of the COVID-19 pandemic in China to be after 20 March 2020, while about 52,000 to 68,000 infected and 2400 death cases were predicted [14]. We aim to propose fourteen prediction models based on artificial neural networks (ANN) to predict the COVID-19 outbreak for policy makers

Objectives

Methods

Results

Discussion

Conclusion