RF-KDE-QSR Model for Estimating the Scale of Epidemics

Chuwei Liu, Jianping Huang, Siyu Chen, Jiaqi He, Shikang Du, Nan Yin, Chao Zhang, Danfeng Wang

Abstract: Infectious diseases are posing an increasingly serious threat to human society. It is urgent to make a rapid estimate of the scale of outbreaks when the disease information is still unclear in the early stages of the outbreak, so as to buy time for a timely response to infectious diseases and provide reference for the allocation of medical resources and the formulation of control measures. Based on this, this study took the concentrated outbreak of COVID-19 in various cities in China as an example, collected 22 meteorological, social-ecological and population mobility indicators, and established a random forest-kernel density estimation-quantile stepwise regression (RF-KDE-QSR) model to make a preliminary estimate of the daily outbreak scale in cities. The RF model was used for preliminary estimation, and the KDE-QSR model was used for residual correction to correct the prediction results. The evaluation of the prediction accuracy proved the effectiveness of the prediction model. When the RF model was used alone, the R-squared (R2) was 0.82 and the corrected R2 was 0.90. The KDE-QSR model effectively improved the prediction accuracy of the model.