A-LSTM Model for Predicting the Deaths Caused by COVID-19 in U.S.

The epidemic caused by a new type of coronavirus has spread around the world since the end of 2019. As of March 4, 2021, over 114.65 million were infected and more than 2.55 million died, including 28.4 million confirmed cases with 0.51 million deaths in U.S. To predict the next 7-day deaths of COVID-19 in U.S. more accurately, we propose a neural network based on LSTM model with attention mechanism. Driven by the historical data provided by the COVID tracking project supported by the Atlantic, the A-LSTM model makes its prediction and its evaluation indexes of RMSE, MAPE, MAE and R-squared are 285.89, 0.0482%, 230.74 and 0.9954 respectively, which are better than the BPNN model's. The result shows that A-LSTM model we propose has a better prediction on the deaths of COVID-19 in U.S. than BPNN.