Front. Res. Metr. Anal.
Sec. Research Policy and Strategic Management
doi: 10.3389/frma.2022.1003972

Modelling COVID-19 incidence with Google Trends

 Lateef B. Amusa1*, Hossana Twinomurinzi1 and Chinedu W. Okonkwo1
  • 1Centre for Applied Data Science, University of Johannesburg, South Africa
Provisionally accepted:
The final, formatted version of the article will be published soon.

Infodemiologic methods could be used to enhance modelling infectious diseases such as COVID-19. We used Google trends data to track COVID-19 incidences and assessed whether they could complement traditional data based solely on reported case numbers. Data on the Nigerian weekly COVID-19 cases spanning through March 1, 2020, to May 31, 2021, were matched with internet search data from Google Trends. The reported weekly incidence numbers and the GT data were split into training and testing sets. ARIMA models were fitted to describe reported weekly COVID cases using the training set. Several COVID-related search terms were theoretically and empirically assessed for initial screening. The utilized Google Trends (GT) variable was added to the ARIMA model as a regressor. Model forecasts, both with and without GTD, were compared with weekly cases in the test set over 13 weeks. Forecast accuracies were compared visually and using RMSE (root mean square error) and MAE (mean average error). Statistical significance of the difference in predictions was determined with the two-sided Diebold-Mariano test. Preliminary results of contemporaneous correlations between COVID-related search terms and weekly COVID cases reveal 'loss of smell', 'loss of taste', 'fever' (in order of magnitude) as significantly associated with the official cases. Predictions of the ARIMA model using solely reported case numbers resulted in an RMSE (root mean squared error) of 411.4 and mean absolute error (MAE) of 354.9. The GT expanded model achieved better forecasting accuracy (RMSE: 388.7 and MAE = 340.1). Corrected Akaike Information Criteria also favoured the GT expanded model (869.4 vs 872.2). The difference in predictive performances was significant when using a two-sided Diebold-Mariano test (DM = 6.75, p < 0.001) for the thirteen weeks. Google trends data enhanced the predictive ability of a traditionally based model and should be considered a suitable method to enhance infectious disease modelling.

Keywords: big data, Google Trends, ARIMA, COVID - 19, infectious disease modeling

Received: 26 Jul 2022; Accepted: 30 Aug 2022.

Copyright: © 2022 Amusa, Twinomurinzi and Okonkwo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence: Dr. Lateef B. Amusa, University of Johannesburg, Centre for Applied Data Science, Johannesburg, South Africa