A Rolling-Based Multipeak Learning Model for COVID-19 Pandemic Predicting

The COVID-19 pandemic has led to a dramatic loss of human life and the global economy, and presents an unprecedented challenge to public health management for all countries around the world. Access to an accurate epidemic prediction model plays a crucial role in epidemic prevention, infection scale control, and medical resource allocation. In this paper, we first propose a multipeak SEIYAQURD model by using the multipeak learning algorithm to predict the COVID-19 epidemic. The model separates the total population according to characteristics of COVID-19 and can capture trend changes in the epidemic. Then, the fitting period technique and the rolling prediction strategy are proposed to improve the prediction accuracy. Numerical experiments based on the data of COVID-19 in the United States are performed to demonstrate the effectiveness of our proposed method by comparing with two benchmark methods from the literature in two cases, one has a smooth trend and the other has a significant changing trend.