Application of an Artificial Neural Network to Classify countries with COVID-19 adjusting to SIR model

Resumen

A backpropagation artificial neural network was employed in MathWorks MATLAB software to analyze the data of 96 countries infected with COVID-19. They were evaluated with the fitVirusCV19v3 (COVID-19 SIR model) function from MathWorks to classify them in two classes. The first-class has an R-square precision greater than 0.995 when evaluating the SIR model and the other with precision less than 0.995 to zero.
The data of the first class has only one maximum and can predict the final of the epidemic with the SIR model. The data of the second class has more than one peak, and it is not possible to find one maximum to evaluate the final of the disease in that country.
To training the artificial neural network, 70% of the data were used, 15% of data were taken for validation, and 15% to testing. In the testing, 5 to 7 countries were well classified and 2 of 7 were classified incorrectly. Then, the model has 75% of precision obtained with the confusion matrix, and the ROC curve shows sufficient success in the classification.
This first approximation to classify an epidemiologic phenomenon through an ANN with the SIR, the model could help to health professionals to know a possible panoramic about the infection course related to sickness, considering the mathematical conditions of the SIR model.