Article Preview
TopIntroduction
World Health Organization (WHO) named the disease caused by novel Coronavirus SARS-CoV-2 as COVID-19 on February 11, 2020 (Li et al., 2020; Dong et al., 2020; Law et al., 2020; Hamid et al., 2020). COVID-19 has initially started with strange new pneumonia of unknown cause in Wuhan City, Hubei province of China (Hui et al, 2020) on December 31, 2019 (WHO, 2020; Sarkodie & Owusu, 2020; Shi et al., 2020). On March 11, 2020, the WHO declared COVID-19 outbreak as pandemic as it widely spreads and infects many people all over the world (WHO, 2020). The entire trajectory of COVID-19 inflammation is growing rapidly in all the countries but awareness about the disease is very low among the population. Many countries are quarantining the affected peoples to reduce the cause of spread and even some people do not know their infection because of no such large number of testing laboratories available in all cities and provinces.
In India, the first case was reported on 30th January 2020 that the patient has travel history from some other country (ICMR, 2020). On 4th March 2020, India has a sudden hike in the increase of infectious rate as then the rate increases very slowly day by day. On 29th March 2020, it crosses more than a thousand in the infectious rate and still, it is increasing in large number. On 27th June 2020, the total number of infected/ confirmed cases is about 67,3165 as per the Indian Council of Medical Research (ICMR) record (ICMR, 2020). Moreover, many countries as like India has more complications in handling COVID-19 due to an increase in globalization, population density in cities, non-compliance of supply and demand of medical facilities, drugs and shortcoming of medical equipment and protections. However, in this complicated pandemic situation forecasting of numbers ahead of month or bi-month may be a significant and alternate strategy to prevent and control the infections. To predict the temporal order of infectious rate, this paper proposes a supervised Long-Short Term Memory (LSTM) model and its variants to predict the number of possible infectious cases ahead of 7, 14 days and a month. The publicly available dataset has been used to predict the infectious rate through the variants of proposed supervised LSTM models such as LSTM, Bi-LSTM and Stacked LSTM.
This kind of infection rate prediction has been analyzed recently in previous studies for COVID-19 pandemic situations. However, the proposed studies mostly follow only a data driven approach and those methods are linear in nature that often neglects the temporal components of the data. Statistical methods such as Auto Regressive Moving Average (ARIMA), Moving Average (MA), Auto Regressive (AR) methods, GARCH models tremendously depends on assumptions and such models are also difficult for forecasting real transmission of COVID-19 epidemic situation (Tomar and Gupta, 2020; Elmousalami and Hassanien, 2020). Certain other studies (Chimmula & Zhang, 2020 and Ayyoubzadeh et al, 2020) have proposed deep learning models for the analysis of infectious cases and death rates using traditional training and testing rather than prediction. Moreover, these studies have better inferences in prediction for shorter periods with minimum loss. In case of longer prediction for a month or above, these technique produces more error.
To analyze the epidemic level for a shorter to longer duration this paper proposes an LSTM model and its variants using one-step prediction with windowing technique. The proposed LSTM model uses R0 reproductive number - a statistical method to infer the prediction of an infectious cause for a month in India. The empirical analysis of infectious case prediction mainly depends on the past historical data collected. However, the COVID-19 epidemic depends on a large number of external and global factors. So this paper suggests a recommendation to all Indian individuals about the awareness and cause of COVID-19 ahead of the month using various analyses.