Probabilistic forecasting of hand, foot and mouth disease in Mainland China using Bayesian additive regression tree model.

Hand, foot, and mouth disease (HFMD) is a serious public health concern in China. Accurately forecasting HFMD and providing forecast uncertainty is significant for decision making by public health workers. However, existing studies have mainly focused on the accuracy of point predictions, with limited research using probabilistic predictions to provide uncertainty. In this study, we used Bayesian additive regression tree probabilistic model for point forecasting and providing forecasting intervals and compared the performance with ARIMA model based on the monthly number of HFMD cases in seven regions of mainland China from June 2008 to December 2018. Compared with ARIMA model, the mean values of MAPE and RMSE of BART model were 42.463 and 4124.997, which decreased by 73.465% and 16.332%, and the mean value of PCC was 0.921, which improved by 10.432%. The interval prediction results showed that the BART model had the smallest CWC values in study areas, covering the actual values with a narrower interval at the 95% confidence level. The BART probabilistic model is suitable for HFMD surveillance at the provincial level in mainland China because of its high point and interval prediction accuracy and strong generalization ability, which helps public health workers in decision making.