Predicting the epidemic curve of the disease COVID-19

In this web page we present COVID-19 prediction curves for the number of new patients based on official WHO data regarding several countries with a daily update. Our prediction model considers a Recurrent Neural Network (RNN) model as an Artificial Intelligence (AI) tool. For more information see our corresponding research paper.

We consider two types of predictions for each country that can be interpreted as follows:

  • Prediction 1: An algorithm to update training step and subsequent prediction was formulated. This update step is based on the general recommendations of transfer learning that considers the already known time interval for the given country and re-training is done in small increments of the RNN network accordingly. Thus, we start predicting the first unknown element x(t+1) from the last 5% of the known data, and the same principle is applied to each subsequent element. Moreover, after each prediction step our RNN architecture is re-trained and the subsequent elements are predicted with this updated RNN.
  • Prediction 2:  We start predicting the first unknown element x(t+1) from the last known x(t), and all the subsequent elements are predicted only from the preceding ones. Here the rules depicted from the training data set are used, not retraining occurs.

The intuitive interpretations of the difference between Prediction 1 and Prediction 2 are as follows. Prediction 2 makes its predictions utilizes the information derived from the training data set, reflective of the trends in the average time series. It follows that predictions will comply primarily with the Hubei time series, especially in the far future. Therefore Prediction 2 shows highest fidelity to the country-specific future scenario if the approach to mitigate the epidemic is similar to that in Hubei. Accordingly, this scenario is also reflective of a country-specific future state given the practices of Hubei were followed in said country. On the other hand, Prediction 1 is yielded after the neural network is retrained after any prediction, providing more valid insight into what is expected if the country goes on with the mitigation practices seen during the observation period.

Predictions for individual countries:

 

 

 

 

 

Related work:

Predicting the epidemic curve of the coronavirus (SARS-CoV-2) disease (COVID-19) using artificial intelligence
 

AUTHORS

Dr. László R. Kolozsvári, PhD,1 Dr. Tamás Bérczes, PhD,2 Dr. András Hajdu, DSc,2 Dr.

Rudolf Gesztelyi, PhD,3 Attila Tiba, MSc,2 Dr. Imre Varga, PhD,2 Gergő J. Szőllősi, MSc,1

Szilvia Harsányi, MSc,4 Dr. Szabolcs Garbóczy,5 MD, Dr. Judit Zsuga, PhD4

  1. Department of Family and Occupational Medicine, Faculty of Public Health, University of Debrecen
    Address: Móricz Zs. krt. 22, 4032, Debrecen, Hungary
  2. Faculty of Informatics, University of Debrecen
    Address: Kassai út 26., 4028 Debrecen, Hungary.
  3. Department of Pharmacology and Pharmacotherapy, Faculty of Medicine, University of Debrecen
    Address: Nagyerdei krt 98., 4032 Debrecen, Hungary
  4. Department of Health Systems Management and Quality Management in Health Care, Faculty of Public Health, University of Debrecen
    Address: Nagyerdei krt 98, 4032 Debrecen, Hungary
  5. Department of Psychiatry, Kenézy Hospital, University of Debrecen,
    Address: Bartok st. 2-26., 4031 Debrecen, Hungary

Preprint is available here.

Last update: 2023. 02. 01. 09:16