Predicting trend of measles epidemic in Hebei province: an empirical study with long short-term memory neural network model

  Objective  To explore the feasibility of predicting the trend of measles epidemic using long short-term memory (LSTM) neural network model for conducting prevention and control of measles.

  Methods  The data on 51 012 measles cases reported in Hebei province form 2004 through 2020 were collected from China Information System for Disease Control and Prevention. The LSTM neural network model was constructed and the optimal model was selected to predict the trend of measles epidemic in the province. Rooted mean squared error (RMSE) and mean absolute error (MAE) were used to evaluate the prediction of model established.

  Results  The annual number of measles cases reported in the province during the 17-year period were 950, 4 837, 7 953, 4 973, 2 273, 3 359, 14 457, 79, 38, 353, 5 365, 3 825, 1 825, 287, 241, 130, and 67, respectively, with a persistent decline since 2015. In addition, an obvious seasonality was observed in the incidence of measles. Using the collected data of 2017, the window length of 3 was determined for the constructed LSTM neural network model, with the RMSE of 17.288 and the MAE of 12.334, and the model was adopted to predict monthly number of measles cases from 2017 through 2020. The predicted monthly numbers of measles incidence were basically consistent with the number observed and the values of RMSE and MAE for years of 2017, 2019 and 2020 were all below 10, but the values for 2018 were slightly higher.

  Conclusion  The constructed LSTM neural network model in this study showed a good efficiency in predicting monthly measles incidence in Hebei province and the model could be used in the analysis on measles incidence trend and epidemic risk assessment.