Prediction Of Influenza Trends In China Based On The Internet

Influenza virus caused large mobtality and morbidity to human society.Timely surveillance is critical for its prevention and control.Traditional surveillance methods rely on a hierarchical surveillance system,which often leads to time delays between outbreaks and their publishments.The recent internet-based computational methods,with the advantage of timeliness and low-cost,have played important roles in prevention and control of infectious diseases.The Google Flu Trends,which was developed by Google Company in 2008,has shown its great potential in prediction of influenza trends in more than 20 countries in the world.Although much progress has been made,most work was conducted in developed countries.Only a few works were conducted in China where the influenza trends is much complex than most countries.In this thesis,we used the big data from Internet,mainly the data from social networks and search engine,to predict the influenza trends in China.The first part of the thesis included the work to predict the influenza trends by province in China using the data from Sina Weibo,a social network similar to Twitter.Firstly,a couple of key words related to influenza were selected manually.Then,the weekly number of appearance of these key words in each province in China was collected through Sina Weibo.Correlation analysis shows that these key words correlated moderately with the influenza trends,with the Pearson Correlation Coefficients(PCC)ranging from 0.56 to 0.71.Further,the daily number of appearance of these key words was used to predict the influenza trends by province with the multiple linear regression.The results show that these models based on key words can accurately predict the influenza trends in China.The second part of the thesis included the work to predict the influenza trends in China using the search index in three platforms,i.e.,the Baidu,Haosou and Weibo index.The daily number of searches for the influenza related key words was retrived through the platforms.Then,correlation analysis and regression modeling were conducted as mentioned above.The results show that Baidu index and Haosou index of influenza-related keywords has a stronger correlation with the number of seasonal influenza-like illness(ILI)in China than the Weibo index,although the correlation for these indexes is moderate.Besides,the epidemic period and peak time for the Baidu index and Haosou index is more similar to those of ILI than Weibo index.Regression analysis further shows that the Baidu index and Haosou index could be used to predict ILI more accurately than Weibo index.Finally,incorporation of the historical ILI could significantly improve the performance of the regression model based on the Baidu index and Haosou index.Overall,this thesis demonstated the possibility of using internet data to predict influenza trends in China,which suggests that the internet-based methods could be a useful supplement to the traditional influenza surveillance methods.