Modelling for prediction of the spread and severity of COVID-19 and its association with socioeconomic factors and virus types

Abstract

We report the development of a Weibull based Long-Short-Term-Memory approach (W-LSTM) for the prediction of COVID-19 disease. The W-LSTM model developed in this study, performs better in terms of MSE, R2 and MAPE, as compared to the previously published models, including ARIMA, LSTM and their variations. Using W-LSTM model, we have predicted the beginning and end of the current cycle of COVID-19 in several countries. Performance of the model was validated as satisfactory in 82% of the 50 test countries, while asking for prediction for 10 days beyond the period of training. Accuracy of the above prediction with days beyond training was assessed in comparison with the MAPE that the model gave with cumulative global data. The model was applied to study correlation between the growth of infection and deaths, and a number of effectors that may influence the epidemic. The model identified age groups, trade with China, air traffic, country temperature and CoV-2 virus types as the likely effectors of infection and virulence leading to deaths. The predictors likely to promote or suppress the epidemic were identified. Some of the predictors had significant effect on the shape parameters of Weibull distribution. The model can function on cloud, take inputs in real time and handle large data country wise, at low costs to make predictions dynamically. Such predictions are highly valuable in guiding policy makers, administration and health. Interactive curves generated from the W-LSTM model can be seen at http://collaboration.coraltele.com/covid2/.