The novel Coronavirus-19 (COVID-19) is an infectious disease and it causes serious lung injury. COVID-19 induces human disease, which has killed numerous people around the world. Moreover, the World Health Organization (WHO) declares this virus as a pandemic and all countries attempt to monitor and control it by locking all places. The illness induces respiratory influenza like problems with symptoms such as cold, cough, fever, and the difficulty of breathing in extremely severe cases. COVID-2019 has been viewed as a global pandemic, and a few analyses are being performed using multiple computational methods to predict the possible development of this pestilence. Considering the various conditions and inquiries these numerical models are based on future tendency. Multiple techniques have been proposed that could be helpful in forecasting the spread of COVID-19. Through statistical modeling on the COVID-19 data, we performed linear regression, random forest, ARIMA and LSTMs, to estimate the empirical indication of COVID-19 ailment and intensity in 4 countries (USA, India, Brazil, and Russia), in order to come up with a better validation.
HIGHLIGHTS
Provide a comparative analysis taking into consideration the dataset from Our World in Data of 4 countries (Brazil, India, Russia, and USA), to provide optimum test cases and validation across multiple trend patterns focused at 2 forecasting events namely, cases forecasting and deaths forecasting
Three statistical modelling approaches (linear regression, random forest, and ARIMA) and 1 deep learning approach (LSTMs) were explored
Performance metrics and diagnostic tools such as residuals, correlograms, RMSE, AIC and BIC were implemented to monitor models’ accuracy
ARIMA outperformed linear regression and random forest in terms of accuracy prediction of test data
ARIMA and LSTMs were compared again with death forecasting task, in which LSTMs were able to provide very high accuracy in comparison
GRAPHICAL ABSTRACT