A Comparative Study On Prediction Models Of Hand Foot And Mouth Disease Based On Baidu Index

ObjectiveSince 2008,hand foot and mouth disease(HFMD)has been included in the statutory group C infectious diseases,the high incidence of HFMD in Qingyuan City has brought a large disease burden.This study intends to use baidu search engine to get the search amount of hand,foot and mouth disease related keywords in a certain period of Qingyuan search volume from the search engine baidu,and use the baidu Index of the keyword to build the prediction model of hand,foot and mouth disease in Qingyuan City,than choosing the model which is the most accurate warning of the incidence of hand,foot and mouth disease in Qingyuan City.The aim is to explore the possibility of establishing a prediction model of HFMD based on Baidu Index in Qingyuan area.It can dynamically monitor the occurrence and development of HFMD in Qingyuan city,and also can early warn the relevant departments to prepare good material,and manpower preparation.Methods(1)Collecting weekly incidence of hand foot mouth disease from 1/1/2013 to 5/31/2017 in the city of Qingyuan from China Disease Prevention and Control Information System,a total of 231 weeks.(2)According to the public search behavior and the incidence and definition of hand foot mouth disease,than combined with the keyword mining tool webmaster tools(http://www.7c.com/keyword/)to select the most representative keywords.The final 15 key words are adopted,there are "Fever","Rash","Herpes","Maculopapule","Enterovirus","Loss of appetite","Mouth ulcers","What to eat for HFMD","HFMD symptoms","HFMD is serious","HFMD in children","HFMD","HFMD picture","Initial symptoms of HFMD","How to treat HFMD".Then collecting the search volume of 15 keywords in the range of Qingyuan city from 1/1/2013 to 5/31/2017 on the weekly from the website of baidu Index(http://index.baidu.com/).(3)After normal distribution analysis,correlation analysis and cross correlation analysis are used to analyze the correlation between the major factors in the system to determine the Baidu Index with the highest correlation with HFMD.(4)The gray model(1,N),multilayer perceptron neural network model and autoregressive distribution lag model are established by using the incidence of HFMD and the Baidu index in the first 227 weeks,then three models are used to predict the incidence of HFMD after 4 weeks.The model are evaluated by the index of mean absolute percent error.Results(1)After the correlation analysis and cross correlation analysis,finally five HFMD keywords as predictors are included.(2)The average absolute error rate of gray model GM(1,N)is 0.26,The MAPE predicted by the multilayer perceptron neural network model is0.38,and the autoregressive distribution lag model is 0.10.Conclusion(1)The fitting effect of gray model GM(1,N)and multilayer perceptron neural network model based on baidu index are not as effective as the autoregressive distribution hysteresis model,and the epidemic situation of HFMD in Qingyuan city remains to be studied.(2)Autoregressive Distributed Lag Model using Baidu Index to predict the incidence of HFMD in Qingyuan city is applicable,but whether it is sustainable and outreach will be the focus of the rest of the work.(3)It is a simple,convenient and low cost method to explore the relationship between the number of key words Baidu index and the number of hand foot and foot disease by Baidu search engine.It provides a new idea and new direction for disease monitoring and prevention and control.It is of great practical value and broad prospect.