Research On ARIMA And BPNN Model In Prediction Of Hand-foot-mouth Disease Incidence

The purpose of this thesis is to analyze the national monthly incidence data of hand-foot-mouth disease from January 2010 to December 2019,and to grasp the prevalence trend of hand-foot-mouth disease in China during the past decade.The seasonal ARIMA model,BPNN model,and ARIMA-BPNN combination model are used to predict the monthly incidence of hand-foot-month disease in China.The quality of these models are compared and estimated,then we combine them with the actual situation of hand-foot-mouth disease in China to establish an appropriate prediction model for the incidence of hand-foot-mouth disease,Aiming to provide reasonable recommendations for the country’s prevention and treatment of hand-foot-mouth disease.We establish a linear model based on the total population of China from 2010 to2018 to estimate the total population of each month.The incidence rate of HFMD in each month from January 2010 to December 2019 are calculated,and their features are analyzed.Then the data from January 2010 to December 2018 are used as the training set to establish the appropriate ARIMA model,BPNN model and ARIMA-BPNN combined model.These models are applied to predict the monthly incidence of hand-foot-mouth disease in China from January to December in 2019,and are compared with real data.Finally,through evaluation indicators,such as mean absolute error（MAE）,mean square error（MSE）,root mean square error（RMSE）and mean absolute percentage error（MAPE）,the advantages and disadvantages of these models are compared.The research results show that the monthly incidence rate of hand-foot-mouth disease in China from 2010 to 2019 is periodic occurring every twelve months.May,June,and July of each year are the periods of highest incidence,and February is the time of lowest incidence.Using the ARIMA model to predict the monthly incidence of hand-foot-mouth disease in China in 2019,the ARIMA（2,0,0）（1,1,1）₁₂model is established.The MAE,MSE,RMSE,and MAPE are 2.12,9.98,3.16,25.13%,respectively.According to the different methods of dividing sample data,three different BPNN models are established.Network1 uses the data of the previous six months to predict the data of the next month,and the best network structure is 6-12-1.Network2uses the data of the previous twelve months to predict the data of the next month,and the best network structure is 12-6-1.Network3 uses the data of the previous three years of the same month to predict the data of the next year in the same month,and the best network structure is 3-8-1.Then we use these models to predict the monthly incidence of hand-foot-mouth disease in China in 2019.The MAE,MSE,RMSE and MAPE of the test set are 1.55,3.49,1.86,19.93%;2.47,10.42,3.22,28.41%;2.33,9.79,3.12,29.4%,respectively.By utilizing training results of the ARIMA model as input data and true values as output data,an ARIMA-BPNN model is established,in which the network structure is 1-8-1.The monthly incidence of hand-foot-mouth disease in China in 2019 can therefore be predicted.The MAE,MSE,RMSE and MAPE of the test set are 1.17,5.62,2.37,15.79%.According to the results of the study,it can be concluded that the monthly incidence of hand-foot-mouth disease in China from 2010 to 2019 shows an obvious seasonal trend,and the number of deaths is gradually decreasing in recent years.Seasonal ARIMA model,BPNN model and ARIMA-BPNN model can all be used to predict the monthly incidence data of hand-foot-mouth disease in China.A total of 5different models have been established in this thesis and the prediction accuracy of these models are compared.The prediction accuracy of ARIMA-BPNN combined model is significantly better than the single ARIMA model and the single BPNN model.