The challenge was to identify patterns between human activity and COVID-19 cases, and to identify factors that can help to predict critical points in the spread of the disease. This has been accomplished. With the neural network created, it was possible to analyze the influence of human factors with the spread of the virus.
Once the critical factors for the spread of COVID-19 have been obtained and bearing in mind that data such as population density, infrastructure, CO2 emissions (to control the circulation of people) and the rate of adherence to social isolation are available from government organizations, as NASA, it was possible to predict the critical points.
For design issues, it was used data provided by IBGE (Brazilian Institute of Geography and Statistics) and websites of local state governments. However, this project aims to be expanded worldwide.
The initial idea was to create an index to list places with higher risks of contagion of COVID-19 using a neural network, but during the process, we were also curious to know which neural network would have a better result (RBF or MLP).
To be able to compare the results of the networks it is necessary to use the same database in both and have a result for comparison that the two networks can generate.
The database we use was built by our team and contains Brazilian data, which we have already mentioned, but this database can be made anywhere in the world using similar metrics and / or adding others, like NASA data. In the case of comparison, we choose the terms "loss" and "r2" to compare the quality of the networks, where "loss" is the metric that measures the cost function (gradient) when calculating synaptic weights (the error ) and "r2" is the correlation coefficient, which determines how much the predicted model looks like the real one (ranges from 0 to 1, positive or negative)
The choice of data was made based on the availability of the data, here in Brazil we don't have data from so many cities, however, the data provided by NASA was not possible to be used due to the formats and how we should work with them, which would take a much longer time than expected in the hackathon just to learn to manipulate this available data. In order to test our idea and assemble the prototype, we chose Brazilian data because of the ease of manipulation.
The neural networks that have been programmed can be accessed through GitHub and the results are also available on the same link.
The results were not what we wanted but we understood that within the limitations that existed, they were acceptable. To improve the accuracy of the networks, it is necessary to make a database more complete than we have and perhaps put new entries in the neural networks (increasing the number of data types, creating a more complete and more accurate analysis).
For a better understanding of the networks used, the next paragraph is focused on their definition.
Radial Basis Function (RBF) is a linear neural network and has its supervised learning (where the network learns from the data set already defined and the inputs and outputs are known). In this type of architecture, there is only an intermediate layer, where the activation functions are of high dimensionality. RBF can be used as a base function for nonlinear regression models (linear or nonlinear) and can also be used as an activation function of any multilayer network type, such as multilayer perceptron (MLP). A Multi Layer Perceptron is a supervised algorithm that can be used for classification or regression. An MLP consists of a three layers: input layers, hidden layer and an output layer. It utilizes a supervised learning technique called backpropagation for training.
http://abes-dn.org.br/pdf/Ranking_2019.pdf (Water accesibility)
https://cidades.ibge.gov.br/ (COVID-19 cases)
https://www.saopaulo.sp.gov.br/coronavirus/isolamento/ (Rate of Adherence to Social Isolation)
https://www.unacast.com/covid19/social-distancing-scoreboard#scoreboard (Reported cases of COVID-19 - US)
https://daac.ornl.gov/cgi-bin/dsviewer.pl?ds_id=1735 (On-road CO2 emissions)
https://www1.folha.uol.com.br/equilibrioesaude/2020/04/veja-em-sua-cidade-quantos-sao-idosos-principal-grupo-de-risco-para-coronavirus.shtml (Rate of Elders per Brazilian city)
http://atlas.ana.gov.br/Atlas/forms/analise/Geral.aspx?est=5(Basic Water Supply info)
https://infosanbas.org.br/ (Complimentary water supply info)
https://mapabrasileirodacovid.inloco.com.br/pt/ (Rate of Adherence to Social Isolation