Analysis of meteorological factors influencing the incidence of influenza in Fujian Province based on a neural network model

Abstract

Abstract Objective: This study aimed to assess and compare the predictive effects of meteorological factors on the incidence of influenza in Fujian Province, China,using four different deep learning network models.Methods: From 2016 to 2020,weekly meteorological and influenza surveillance data in Fujian Province were collected. Using four different deep learning network models, including ordinary neural network (ANN), deep neural network (DNN), recurrent neural network (RNN), and gated recurrent unit (GRU), the prediction model of the weekly average temperature, influenza lag and influenza incidence were determined, and the predictive effects from each different models were compared.Results: The incidence of influenza in Fujian Province showed obvious seasonality, with a high incidence in winter, especially from November to March, during which influenza incidence reached the highest value each year. A non-linear negative correlation between temperature and incidence of influenza was obtained. Compared with the prediction model that only considers “temperature” as a factor, the model that includes both temperature and lag had a better predictive effect. Overall, the GRU model, with three hidden layers (constructed from temperature, influenza lag of one week and two weeks), had the best prediction ability, followed by RNN, DNN, and ANN, respectively.Conclusion: Temperature and influenza incidence showed a non-linear negative correlation. Furthermore, the GRU model provides a better prediction of the influenza incidence and, therefore, can be used to develop an influenza risk early warning system based on temperature and influenza lag, to prevent the incidence and spread of influenza.