Influenza Activity Surveillance Based on Multiple Regression Model and Artificial Neural Network

Abstract

In this paper, a series of models were established, and based on the Google Flu Trends (GFT) data and Centers for Disease Control (CDC) data. The models include the GFT regression model (model 1), the weighted GFT regression model (model 2), the GFT + CDC regression model (model 3), the CDC regression model (model 4), and the weighted CDC regression model (model 5). All models were utilized to predict and assess influenza activity across ten regions of the United States. The least squares and backpropagation neural network based on the genetic algorithm are used to fit the model parameters, and the error and historical sample fitting accuracy of each model are compared. The results show that models 4 and 5 are superior to other models. To optimize the prediction model, the seasonal characteristics of influenza incidence were investigated, and flu-prediction models for the high-flu season and low-flu season were established. The experimental results show that the prediction model of seasonal influenza is superior to the non-seasonal model. The influenza-like illness values predicted by the seasonal flu model are consistent with information provided by the CDC, suggesting that the results accurately reflect influenza epidemic characteristics and can thus be readily applied for the prevention and control of influenza.