Abstract: Objective Constructing an optimal model for varicella prediction in Nanjing City to provide scientific guidance for the prevention and control of varicella. Methods Constructed Holt-Winters addition model, Holt-Winters product model and autoregressive integrated moving average （ARIMA） model based on the monthly varicella incidence rate in Nanjing from 2014 to 2018. By testing the parameters of the models and calculating the relative errors between predicted and actual values, selected the optimal model to predict varicella incidence in Nanjing in 2019. Results Based on Ljung-BOX test level, rejecting the Holt-Winters addition model and selecting the Holt-Winters product model and ARIMA model as the varicella prediction model. The bayesian information rules （BIC） of Holt-winters product model and ARIMA model was 1.24 and 1.81, respectively; the mean absolute percentage error （MAPE） was 14.24% and 21.86% respectively; the R² was 0.97 and 0.97, respectively; the average relative error between the predicted value and the actual value was 11.90% and 15.76%, respectively. Conclusion Holt winters product model is superior to ARIMA model in fitting and forecasting effect, and it is a high accuracy model for varicella prediction in the short term.

Key words: Varicella, Holt-Winters model, ARIMA, Forecast