Real-time forecasts and risk assessment of novel coronavirus (COVID-19) cases: A data-driven analysis

Tanujit Chakraborty; Indrajit Ghosh

To answer the first question, we study classical and modern forecasting techniques for which the prediction accuracy largely depend on the availability of data [28] . In outbreaks of COVID-19 epidemics, there are limited data available, making predictions widely uncertain. From previous studies, it was evident that the timing and location of the outbreak facilitated the rapid transmission of the virus within a highly mobile population [29] . In most of the affected countries, the governments implemented a strict lockdown in subsequent days of initial transmission of the virus and within hospitals, patients who fulfill clinical and epidemiological characteristics of COVID-19 are immediately isolated. The constant increase in the global number of COVID-19 cases is putting a substantial burden on the health care system for Canada, France, India, South Korea, and the UK. To anticipate additional resources to combat the epidemic, various mathematical and statistical forecasting tools [21; 34] and outside China [20; 36; 10] were applied to generate short-term and long-term forecasts of reported cases. These model predictions have shown a wide range of variations. Since the time series datasets of COVID-19 contain both nonlinear and nonstationary patterns, therefore, making decisions based on an individual model would be critical. In this study, we propose a hybrid modeling approach to generate short-term forecasts for multiple countries. In traditional time series forecasting, the autoregressive integrated moving average (ARIMA) model is used predominantly for forecasting linear time series [6] . But in recent literature, the wavelet transformation based forecasting model has shown excellent performance in nonstationary time series data modeling [27] . Thus, combining both models may accurately model such complex autocorrelation structures in the COVID-19 time-series datasets and reduce the bias and variances of the prediction error of the component models. In the absence of vaccines or antiviral drugs for COVID-19, these estimates will provide an insight into the resource allocations for the exceedingly affected countries to keep this epidemic under control. Besides shedding light on the dynamics of COVID-19 spreading, the practical intent of this data-driven analysis is to provide government officials with realistic estimates for the magnitude of the epidemic for policy-making.