Prediction for the Second Wave of COVID-19 in India

Abstract

COVID-19 has been declared as a global pandemic by World Health Organization on $$11^{th}$$ March 2020. Following the subsequent stages of unlocking by the Indian Government, the active cases in India are rapidly increasing everyday. In this context, this paper carries out the comprehensive study of active COVID-19 cases in four states of India namely, Maharashtra, Kerala, Gujarat and Delhi, and predicts the arrival of second wave of COVID-19 in India. Further, since the number of cases reported varies significantly, we utilize the Multiplicative Long-short term memory (M-LSTM) architecture for predicting the second wave. In our experiment, multi-step prediction method is utilized to forecast the active cases for next six months in the four states. Since the input instances vary abruptly with time, simple Long-short term memory (LSTM) is not efficient enough to predict future instances accurately. Our results reveal that M-LSTM have outperformed simple LSTM in predicting the cases. The percentage of improvement of M-LSTM model as compared to simple LSTM is 22.3%. The error rate calculated in terms of N-RMSE (Normalized Root Mean Square Error) for M-LSTM is less than that of each state’s LSTM model. A nested cross-validation method known as Day Forward Chaining improves both models’ performance and avoids biased prediction errors. This technique helped in accurately predicting the active cases by degrading the error values. Our work can help the government and medical officials to better organize their policies and to prepare in advance for increase in the requirement of healthcare workers, medicines and support systems in controlling the upcoming COVID-19 situation.