Analysis and prediction of novel coronavirus pneumonia epidemic using hybrid response surface method with time-series and random forest

Abstract

This paper proposes a new epidemic prediction model that hybridizes several models, such as the autoregressive integrated moving average model (ARIMA), random forest (RF), and response surface method (RSM). The modeling process based on ensemble empirical mode decomposition (EEMD) is particularly suitable for dealing with non-stationary and nonlinear data. ARIMA’s timeliness and difference have strong deterministic information extraction ability. RF is robust and stable, with fast speed, and strong generalization ability. Under the adjustability and correspondence of the response surface, the comprehensiveness of the model is well demonstrated. Taking the United States as an example, the proposed ARIMA-RF-RSM model is used to explore the development mechanism of the early epidemic according to the data of the early epidemic of coronavirus disease 2019 (COVID-19). The proposed model has high prediction accuracy (mean absolute percentage error (MAPE) is 1.97% and root mean square error (RSME) is 7.24%). It helps to take effective prevention and control measures in time. In addition, the model has universal applicability to the analysis of disease transmission in relevant areas.