Forecasting the Number of Daily Covid-19 Cases using Recurrent Neural Networks
The current worldwide outbreak of COVID-19 has presented numerous challenges to the research community. Forecast of the number of patients likely to be infected in a particular region can be of immense help to policy-makers, healthcare organizations and governments. Neural networks have been shown to be successful in a wide variety of problems such as predicting the parameters, risk and effects of an epidemic. However, these models find it difficult to make accurate predictions in low-data environment.
In the present work, we show that a Long Short Term Memory based Recurrent Neural Network can be used to forecast the number of daily Covid-19 cases per country. The results show that our method can outperform state of the art models for 170 countries’ data. This model can be widely used by governments, health organizations, policy-makers and other stakeholders.