COVID-19 positive cases prediction based on LSTM algorithm and its variants

In this paper, deep learning methods are applied to predict positive cases reported in Wuhan and four states in USA. Recurrent neural network based on long-short term memory (LSTM) and its variants including bidirectional LSTM, stacked LSTM and traditional SEIR model are applied on Wuhan dataset to compare and select the best model in task of predicting positive cases. The results reveal that our models based on LSTM significantly perform better than traditional SEIR model. Besides, since bidirectional LSTM can learn information from history and future, it achieves the highest prediction accuracy. Then we use bidirectional LSTM to make prediction on another USA dataset, which contains more recent data. The bidirectional LSTM shows its power and accuracy on this data, which demonstrates its effectiveness on predicting COVID-19 positive cases once again. The model we proposed alos provide some insight into the research of epidemics and the understanding of the spread of the COVID-19.