Covid-19 Epidemic Trend Prediction Based on CNN-StackBiLSTM

In late 2019, the novel coronavirus (COVID-19) became a major health hazard around the world. Recently, the COVID-19 has spread widely in most countries and regions and the number of infected people continues to grow rapidly. Therefore, it is essential to research the development trend of the epidemic. The prediction of the number of infections and deaths is critical and helpful for developing health and epidemic prevention strategies and even curbing the epidemic. In this paper, a one-dimensional convolutional neural network combined with the stacked long-short-term-memory network model (CNN-StackBiLSTM) is proposed for the time-series prediction of cumulative cases and daily new cases. The local feature is extracted by CNN. The stacked BiLSTM captures the deeper characteristics of the time-series data. By combining the two networks, the proposed method simultaneously considers the information of temporal and spatial domains and can achieve accurate prediction results. Examples in Taiwan and Italy demonstrate the effectiveness of the proposed method. The proposed method is compared with LSTM, BiLSTM, and GRU. The mean absolute error, mean square error, R2 score, and root mean square error are calculated to quantificationally measure the different models. The results indicate the proposed method performs well in the prediction of both new daily confirmed cases and cumulative confirmed cases.